xclele/yolov10
1
0
Fork 0
mirror of https://github.com/THU-MIG/yolov10.git synced 2025-05-23 05:24:22 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

x

Browse Source
This commit is contained in:
cq-lsy 2025-03-18 11:22:36 +08:00
parent 84128a569e
commit 1385285579
196 changed files with 0 additions and 38016 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
ultralytics/__init__.py
View File

@ -1,27 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
__version__ = "8.1.34"
from ultralytics.data.explorer.explorer import Explorer
from ultralytics.models import RTDETR, SAM, YOLO, YOLOWorld, YOLOv10,YOLOv10Seg
from ultralytics.models.fastsam import FastSAM
from ultralytics.models.nas import NAS
from ultralytics.utils import ASSETS, SETTINGS as settings
from ultralytics.utils.checks import check_yolo as checks
from ultralytics.utils.downloads import download
__all__ = (
"__version__",
"ASSETS",
"YOLO",
"YOLOWorld",
"NAS",
"SAM",
"FastSAM",
"RTDETR",
"checks",
"download",
"settings",
"Explorer",
"YOLOv10"
)

613
ultralytics/cfg/__init__.py
View File

@ -1,613 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import contextlib
import os
import shutil
import subprocess
import sys
from pathlib import Path
from types import SimpleNamespace
from typing import Dict, List, Union
import re
from ultralytics.utils import (
ASSETS,
DEFAULT_CFG,
DEFAULT_CFG_DICT,
DEFAULT_CFG_PATH,
LOGGER,
RANK,
ROOT,
RUNS_DIR,
SETTINGS,
SETTINGS_YAML,
TESTS_RUNNING,
IterableSimpleNamespace,
__version__,
checks,
colorstr,
deprecation_warn,
yaml_load,
yaml_print,
)
# Define valid tasks and modes
MODES = {"train", "val", "predict", "export", "track", "benchmark"}
TASKS = {"detect", "segment", "classify", "pose", "obb"}
TASK2DATA = {
"detect": "coco8.yaml",
"segment": "coco8-seg.yaml",
"classify": "imagenet10",
"pose": "coco8-pose.yaml",
"obb": "dota8.yaml",
}
TASK2MODEL = {
"detect": "yolov8n.pt",
"segment": "yolov8n-seg.pt",
"classify": "yolov8n-cls.pt",
"pose": "yolov8n-pose.pt",
"obb": "yolov8n-obb.pt",
}
TASK2METRIC = {
"detect": "metrics/mAP50-95(B)",
"segment": "metrics/mAP50-95(M)",
"classify": "metrics/accuracy_top1",
"pose": "metrics/mAP50-95(P)",
"obb": "metrics/mAP50-95(B)",
}
CLI_HELP_MSG = f"""
Arguments received: {str(['yolo'] + sys.argv[1:])}. Ultralytics 'yolo' commands use the following syntax:
yolo TASK MODE ARGS
Where TASK (optional) is one of {TASKS}
MODE (required) is one of {MODES}
ARGS (optional) are any number of custom 'arg=value' pairs like 'imgsz=320' that override defaults.
See all ARGS at https://docs.ultralytics.com/usage/cfg or with 'yolo cfg'
1. Train a detection model for 10 epochs with an initial learning_rate of 0.01
yolo train data=coco128.yaml model=yolov8n.pt epochs=10 lr0=0.01
2. Predict a YouTube video using a pretrained segmentation model at image size 320:
yolo predict model=yolov8n-seg.pt source='https://youtu.be/LNwODJXcvt4' imgsz=320
3. Val a pretrained detection model at batch-size 1 and image size 640:
yolo val model=yolov8n.pt data=coco128.yaml batch=1 imgsz=640
4. Export a YOLOv8n classification model to ONNX format at image size 224 by 128 (no TASK required)
yolo export model=yolov8n-cls.pt format=onnx imgsz=224,128
6. Explore your datasets using semantic search and SQL with a simple GUI powered by Ultralytics Explorer API
yolo explorer
5. Run special commands:
yolo help
yolo checks
yolo version
yolo settings
yolo copy-cfg
yolo cfg
Docs: https://docs.ultralytics.com
Community: https://community.ultralytics.com
GitHub: https://github.com/ultralytics/ultralytics
"""
# Define keys for arg type checks
CFG_FLOAT_KEYS = {"warmup_epochs", "box", "cls", "dfl", "degrees", "shear", "time"}
CFG_FRACTION_KEYS = {
"dropout",
"iou",
"lr0",
"lrf",
"momentum",
"weight_decay",
"warmup_momentum",
"warmup_bias_lr",
"label_smoothing",
"hsv_h",
"hsv_s",
"hsv_v",
"translate",
"scale",
"perspective",
"flipud",
"fliplr",
"bgr",
"mosaic",
"mixup",
"copy_paste",
"conf",
"iou",
"fraction",
} # fraction floats 0.0 - 1.0
CFG_INT_KEYS = {
"epochs",
"patience",
"batch",
"workers",
"seed",
"close_mosaic",
"mask_ratio",
"max_det",
"vid_stride",
"line_width",
"workspace",
"nbs",
"save_period",
}
CFG_BOOL_KEYS = {
"save",
"exist_ok",
"verbose",
"deterministic",
"single_cls",
"rect",
"cos_lr",
"overlap_mask",
"val",
"save_json",
"save_hybrid",
"half",
"dnn",
"plots",
"show",
"save_txt",
"save_conf",
"save_crop",
"save_frames",
"show_labels",
"show_conf",
"visualize",
"augment",
"agnostic_nms",
"retina_masks",
"show_boxes",
"keras",
"optimize",
"int8",
"dynamic",
"simplify",
"nms",
"profile",
"multi_scale",
}
def cfg2dict(cfg):
"""
Convert a configuration object to a dictionary, whether it is a file path, a string, or a SimpleNamespace object.
Args:
cfg (str | Path | dict | SimpleNamespace): Configuration object to be converted to a dictionary.
Returns:
cfg (dict): Configuration object in dictionary format.
"""
if isinstance(cfg, (str, Path)):
cfg = yaml_load(cfg) # load dict
elif isinstance(cfg, SimpleNamespace):
cfg = vars(cfg) # convert to dict
return cfg
def get_cfg(cfg: Union[str, Path, Dict, SimpleNamespace] = DEFAULT_CFG_DICT, overrides: Dict = None):
"""
Load and merge configuration data from a file or dictionary.
Args:
cfg (str | Path | Dict | SimpleNamespace): Configuration data.
overrides (str | Dict | optional): Overrides in the form of a file name or a dictionary. Default is None.
Returns:
(SimpleNamespace): Training arguments namespace.
"""
cfg = cfg2dict(cfg)
# Merge overrides
if overrides:
overrides = cfg2dict(overrides)
if "save_dir" not in cfg:
overrides.pop("save_dir", None) # special override keys to ignore
check_dict_alignment(cfg, overrides)
cfg = {**cfg, **overrides} # merge cfg and overrides dicts (prefer overrides)
# Special handling for numeric project/name
for k in "project", "name":
if k in cfg and isinstance(cfg[k], (int, float)):
cfg[k] = str(cfg[k])
if cfg.get("name") == "model": # assign model to 'name' arg
cfg["name"] = cfg.get("model", "").split(".")[0]
LOGGER.warning(f"WARNING ⚠️ 'name=model' automatically updated to 'name={cfg['name']}'.")
# Type and Value checks
check_cfg(cfg)
# Return instance
return IterableSimpleNamespace(**cfg)
def check_cfg(cfg, hard=True):
"""Check Ultralytics configuration argument types and values."""
for k, v in cfg.items():
if v is not None: # None values may be from optional args
if k in CFG_FLOAT_KEYS and not isinstance(v, (int, float)):
if hard:
raise TypeError(
f"'{k}={v}' is of invalid type {type(v).__name__}. "
f"Valid '{k}' types are int (i.e. '{k}=0') or float (i.e. '{k}=0.5')"
)
cfg[k] = float(v)
elif k in CFG_FRACTION_KEYS:
if not isinstance(v, (int, float)):
if hard:
raise TypeError(
f"'{k}={v}' is of invalid type {type(v).__name__}. "
f"Valid '{k}' types are int (i.e. '{k}=0') or float (i.e. '{k}=0.5')"
)
cfg[k] = v = float(v)
if not (0.0 <= v <= 1.0):
raise ValueError(f"'{k}={v}' is an invalid value. " f"Valid '{k}' values are between 0.0 and 1.0.")
elif k in CFG_INT_KEYS and not isinstance(v, int):
if hard:
raise TypeError(
f"'{k}={v}' is of invalid type {type(v).__name__}. " f"'{k}' must be an int (i.e. '{k}=8')"
)
cfg[k] = int(v)
elif k in CFG_BOOL_KEYS and not isinstance(v, bool):
if hard:
raise TypeError(
f"'{k}={v}' is of invalid type {type(v).__name__}. "
f"'{k}' must be a bool (i.e. '{k}=True' or '{k}=False')"
)
cfg[k] = bool(v)
def get_save_dir(args, name=None):
"""Return save_dir as created from train/val/predict arguments."""
if getattr(args, "save_dir", None):
save_dir = args.save_dir
else:
from ultralytics.utils.files import increment_path
project = args.project or (ROOT.parent / "tests/tmp/runs" if TESTS_RUNNING else RUNS_DIR) / args.task
name = name or args.name or f"{args.mode}"
save_dir = increment_path(Path(project) / name, exist_ok=args.exist_ok if RANK in (-1, 0) else True)
return Path(save_dir)
def _handle_deprecation(custom):
"""Hardcoded function to handle deprecated config keys."""
for key in custom.copy().keys():
if key == "boxes":
deprecation_warn(key, "show_boxes")
custom["show_boxes"] = custom.pop("boxes")
if key == "hide_labels":
deprecation_warn(key, "show_labels")
custom["show_labels"] = custom.pop("hide_labels") == "False"
if key == "hide_conf":
deprecation_warn(key, "show_conf")
custom["show_conf"] = custom.pop("hide_conf") == "False"
if key == "line_thickness":
deprecation_warn(key, "line_width")
custom["line_width"] = custom.pop("line_thickness")
return custom
def check_dict_alignment(base: Dict, custom: Dict, e=None):
"""
This function checks for any mismatched keys between a custom configuration list and a base configuration list. If
any mismatched keys are found, the function prints out similar keys from the base list and exits the program.
Args:
custom (dict): a dictionary of custom configuration options
base (dict): a dictionary of base configuration options
e (Error, optional): An optional error that is passed by the calling function.
"""
custom = _handle_deprecation(custom)
base_keys, custom_keys = (set(x.keys()) for x in (base, custom))
mismatched = [k for k in custom_keys if k not in base_keys]
if mismatched:
from difflib import get_close_matches
string = ""
for x in mismatched:
matches = get_close_matches(x, base_keys) # key list
matches = [f"{k}={base[k]}" if base.get(k) is not None else k for k in matches]
match_str = f"Similar arguments are i.e. {matches}." if matches else ""
string += f"'{colorstr('red', 'bold', x)}' is not a valid YOLO argument. {match_str}\n"
raise SyntaxError(string + CLI_HELP_MSG) from e
def merge_equals_args(args: List[str]) -> List[str]:
"""
Merges arguments around isolated '=' args in a list of strings. The function considers cases where the first
argument ends with '=' or the second starts with '=', as well as when the middle one is an equals sign.
Args:
args (List[str]): A list of strings where each element is an argument.
Returns:
(List[str]): A list of strings where the arguments around isolated '=' are merged.
"""
new_args = []
for i, arg in enumerate(args):
if arg == "=" and 0 < i < len(args) - 1: # merge ['arg', '=', 'val']
new_args[-1] += f"={args[i + 1]}"
del args[i + 1]
elif arg.endswith("=") and i < len(args) - 1 and "=" not in args[i + 1]: # merge ['arg=', 'val']
new_args.append(f"{arg}{args[i + 1]}")
del args[i + 1]
elif arg.startswith("=") and i > 0: # merge ['arg', '=val']
new_args[-1] += arg
else:
new_args.append(arg)
return new_args
def handle_yolo_hub(args: List[str]) -> None:
"""
Handle Ultralytics HUB command-line interface (CLI) commands.
This function processes Ultralytics HUB CLI commands such as login and logout.
It should be called when executing a script with arguments related to HUB authentication.
Args:
args (List[str]): A list of command line arguments
Example:
```bash
python my_script.py hub login your_api_key
```
"""
from ultralytics import hub
if args[0] == "login":
key = args[1] if len(args) > 1 else ""
# Log in to Ultralytics HUB using the provided API key
hub.login(key)
elif args[0] == "logout":
# Log out from Ultralytics HUB
hub.logout()
def handle_yolo_settings(args: List[str]) -> None:
"""
Handle YOLO settings command-line interface (CLI) commands.
This function processes YOLO settings CLI commands such as reset.
It should be called when executing a script with arguments related to YOLO settings management.
Args:
args (List[str]): A list of command line arguments for YOLO settings management.
Example:
```bash
python my_script.py yolo settings reset
```
"""
url = "https://docs.ultralytics.com/quickstart/#ultralytics-settings" # help URL
try:
if any(args):
if args[0] == "reset":
SETTINGS_YAML.unlink() # delete the settings file
SETTINGS.reset() # create new settings
LOGGER.info("Settings reset successfully") # inform the user that settings have been reset
else: # save a new setting
new = dict(parse_key_value_pair(a) for a in args)
check_dict_alignment(SETTINGS, new)
SETTINGS.update(new)
LOGGER.info(f"💡 Learn about settings at {url}")
yaml_print(SETTINGS_YAML) # print the current settings
except Exception as e:
LOGGER.warning(f"WARNING ⚠️ settings error: '{e}'. Please see {url} for help.")
def handle_explorer():
"""Open the Ultralytics Explorer GUI."""
checks.check_requirements("streamlit")
LOGGER.info("💡 Loading Explorer dashboard...")
subprocess.run(["streamlit", "run", ROOT / "data/explorer/gui/dash.py", "--server.maxMessageSize", "2048"])
def parse_key_value_pair(pair):
"""Parse one 'key=value' pair and return key and value."""
k, v = pair.split("=", 1) # split on first '=' sign
k, v = k.strip(), v.strip() # remove spaces
assert v, f"missing '{k}' value"
return k, smart_value(v)
def smart_value(v):
"""Convert a string to an underlying type such as int, float, bool, etc."""
v_lower = v.lower()
if v_lower == "none":
return None
elif v_lower == "true":
return True
elif v_lower == "false":
return False
else:
with contextlib.suppress(Exception):
return eval(v)
return v
def entrypoint(debug=""):
"""
This function is the ultralytics package entrypoint, it's responsible for parsing the command line arguments passed
to the package.
This function allows for:
- passing mandatory YOLO args as a list of strings
- specifying the task to be performed, either 'detect', 'segment' or 'classify'
- specifying the mode, either 'train', 'val', 'test', or 'predict'
- running special modes like 'checks'
- passing overrides to the package's configuration
It uses the package's default cfg and initializes it using the passed overrides.
Then it calls the CLI function with the composed cfg
"""
args = (debug.split(" ") if debug else sys.argv)[1:]
if not args: # no arguments passed
LOGGER.info(CLI_HELP_MSG)
return
special = {
"help": lambda: LOGGER.info(CLI_HELP_MSG),
"checks": checks.collect_system_info,
"version": lambda: LOGGER.info(__version__),
"settings": lambda: handle_yolo_settings(args[1:]),
"cfg": lambda: yaml_print(DEFAULT_CFG_PATH),
"hub": lambda: handle_yolo_hub(args[1:]),
"login": lambda: handle_yolo_hub(args),
"copy-cfg": copy_default_cfg,
"explorer": lambda: handle_explorer(),
}
full_args_dict = {**DEFAULT_CFG_DICT, **{k: None for k in TASKS}, **{k: None for k in MODES}, **special}
# Define common misuses of special commands, i.e. -h, -help, --help
special.update({k[0]: v for k, v in special.items()}) # singular
special.update({k[:-1]: v for k, v in special.items() if len(k) > 1 and k.endswith("s")}) # singular
special = {**special, **{f"-{k}": v for k, v in special.items()}, **{f"--{k}": v for k, v in special.items()}}
overrides = {} # basic overrides, i.e. imgsz=320
for a in merge_equals_args(args): # merge spaces around '=' sign
if a.startswith("--"):
LOGGER.warning(f"WARNING ⚠️ argument '{a}' does not require leading dashes '--', updating to '{a[2:]}'.")
a = a[2:]
if a.endswith(","):
LOGGER.warning(f"WARNING ⚠️ argument '{a}' does not require trailing comma ',', updating to '{a[:-1]}'.")
a = a[:-1]
if "=" in a:
try:
k, v = parse_key_value_pair(a)
if k == "cfg" and v is not None: # custom.yaml passed
LOGGER.info(f"Overriding {DEFAULT_CFG_PATH} with {v}")
overrides = {k: val for k, val in yaml_load(checks.check_yaml(v)).items() if k != "cfg"}
else:
overrides[k] = v
except (NameError, SyntaxError, ValueError, AssertionError) as e:
check_dict_alignment(full_args_dict, {a: ""}, e)
elif a in TASKS:
overrides["task"] = a
elif a in MODES:
overrides["mode"] = a
elif a.lower() in special:
special[a.lower()]()
return
elif a in DEFAULT_CFG_DICT and isinstance(DEFAULT_CFG_DICT[a], bool):
overrides[a] = True # auto-True for default bool args, i.e. 'yolo show' sets show=True
elif a in DEFAULT_CFG_DICT:
raise SyntaxError(
f"'{colorstr('red', 'bold', a)}' is a valid YOLO argument but is missing an '=' sign "
f"to set its value, i.e. try '{a}={DEFAULT_CFG_DICT[a]}'\n{CLI_HELP_MSG}"
)
else:
check_dict_alignment(full_args_dict, {a: ""})
# Check keys
check_dict_alignment(full_args_dict, overrides)
# Mode
mode = overrides.get("mode")
if mode is None:
mode = DEFAULT_CFG.mode or "predict"
LOGGER.warning(f"WARNING ⚠️ 'mode' argument is missing. Valid modes are {MODES}. Using default 'mode={mode}'.")
elif mode not in MODES:
raise ValueError(f"Invalid 'mode={mode}'. Valid modes are {MODES}.\n{CLI_HELP_MSG}")
# Task
task = overrides.pop("task", None)
if task:
if task not in TASKS:
raise ValueError(f"Invalid 'task={task}'. Valid tasks are {TASKS}.\n{CLI_HELP_MSG}")
if "model" not in overrides:
overrides["model"] = TASK2MODEL[task]
# Model
model = overrides.pop("model", DEFAULT_CFG.model)
if model is None:
model = "yolov8n.pt"
LOGGER.warning(f"WARNING ⚠️ 'model' argument is missing. Using default 'model={model}'.")
overrides["model"] = model
# stem = Path(model).stem.lower()
stem = model.lower()
if "rtdetr" in stem: # guess architecture
from ultralytics import RTDETR
model = RTDETR(model) # no task argument
elif "fastsam" in stem:
from ultralytics import FastSAM
model = FastSAM(model)
elif "sam" in stem:
from ultralytics import SAM
model = SAM(model)
elif re.search("v3|v5|v6|v8|v9", stem):
from ultralytics import YOLO
model = YOLO(model, task=task)
else:
from ultralytics import YOLOv10
# Special case for the HuggingFace Hub
split_path = model.split('/')
if len(split_path) == 2 and (not os.path.exists(model)):
model = YOLOv10.from_pretrained(model)
else:
model = YOLOv10(model)
if isinstance(overrides.get("pretrained"), str):
model.load(overrides["pretrained"])
# Task Update
if task != model.task:
if task:
LOGGER.warning(
f"WARNING ⚠️ conflicting 'task={task}' passed with 'task={model.task}' model. "
f"Ignoring 'task={task}' and updating to 'task={model.task}' to match model."
)
task = model.task
# Mode
if mode in ("predict", "track") and "source" not in overrides:
overrides["source"] = DEFAULT_CFG.source or ASSETS
LOGGER.warning(f"WARNING ⚠️ 'source' argument is missing. Using default 'source={overrides['source']}'.")
elif mode in ("train", "val"):
if "data" not in overrides and "resume" not in overrides:
overrides["data"] = DEFAULT_CFG.data or TASK2DATA.get(task or DEFAULT_CFG.task, DEFAULT_CFG.data)
LOGGER.warning(f"WARNING ⚠️ 'data' argument is missing. Using default 'data={overrides['data']}'.")
elif mode == "export":
if "format" not in overrides:
overrides["format"] = DEFAULT_CFG.format or "torchscript"
LOGGER.warning(f"WARNING ⚠️ 'format' argument is missing. Using default 'format={overrides['format']}'.")
# Run command in python
getattr(model, mode)(**overrides) # default args from model
# Show help
LOGGER.info(f"💡 Learn more at https://docs.ultralytics.com/modes/{mode}")
# Special modes --------------------------------------------------------------------------------------------------------
def copy_default_cfg():
"""Copy and create a new default configuration file with '_copy' appended to its name."""
new_file = Path.cwd() / DEFAULT_CFG_PATH.name.replace(".yaml", "_copy.yaml")
shutil.copy2(DEFAULT_CFG_PATH, new_file)
LOGGER.info(
f"{DEFAULT_CFG_PATH} copied to {new_file}\n"
f"Example YOLO command with this new custom cfg:\n yolo cfg='{new_file}' imgsz=320 batch=8"
)
if __name__ == "__main__":
# Example: entrypoint(debug='yolo predict model=yolov8n.pt')
entrypoint(debug="")

127
ultralytics/cfg/default.yaml
View File

@ -1,127 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# Default training settings and hyperparameters for medium-augmentation COCO training
task: detect # (str) YOLO task, i.e. detect, segment, classify, pose
mode: train # (str) YOLO mode, i.e. train, val, predict, export, track, benchmark
# Train settings -------------------------------------------------------------------------------------------------------
model: # (str, optional) path to model file, i.e. yolov8n.pt, yolov8n.yaml
data: # (str, optional) path to data file, i.e. coco128.yaml
epochs: 100 # (int) number of epochs to train for
time: # (float, optional) number of hours to train for, overrides epochs if supplied
patience: 100 # (int) epochs to wait for no observable improvement for early stopping of training
batch: 8 # (int) number of images per batch (-1 for AutoBatch)
imgsz: 640 # (int | list) input images size as int for train and val modes, or list[w,h] for predict and export modes
save: True # (bool) save train checkpoints and predict results
save_period: -1 # (int) Save checkpoint every x epochs (disabled if < 1)
val_period: 1 # (int) Validation every x epochs
cache: False # (bool) True/ram, disk or False. Use cache for data loading
device: # (int | str | list, optional) device to run on, i.e. cuda device=0 or device=0,1,2,3 or device=cpu
workers: 2 # (int) number of worker threads for data loading (per RANK if DDP)
project: # (str, optional) project name
name: # (str, optional) experiment name, results saved to 'project/name' directory
exist_ok: False # (bool) whether to overwrite existing experiment
pretrained: True # (bool | str) whether to use a pretrained model (bool) or a model to load weights from (str)
optimizer: auto # (str) optimizer to use, choices=[SGD, Adam, Adamax, AdamW, NAdam, RAdam, RMSProp, auto]
verbose: True # (bool) whether to print verbose output
seed: 0 # (int) random seed for reproducibility
deterministic: True # (bool) whether to enable deterministic mode
single_cls: False # (bool) train multi-class data as single-class
rect: False # (bool) rectangular training if mode='train' or rectangular validation if mode='val'
cos_lr: False # (bool) use cosine learning rate scheduler
close_mosaic: 10 # (int) disable mosaic augmentation for final epochs (0 to disable)
resume: False # (bool) resume training from last checkpoint
amp: True # (bool) Automatic Mixed Precision (AMP) training, choices=[True, False], True runs AMP check
fraction: 1.0 # (float) dataset fraction to train on (default is 1.0, all images in train set)
profile: False # (bool) profile ONNX and TensorRT speeds during training for loggers
freeze: None # (int | list, optional) freeze first n layers, or freeze list of layer indices during training
multi_scale: False # (bool) Whether to use multiscale during training
# Segmentation
overlap_mask: True # (bool) masks should overlap during training (segment train only)
mask_ratio: 4 # (int) mask downsample ratio (segment train only)
# Classification
dropout: 0.0 # (float) use dropout regularization (classify train only)
# Val/Test settings ----------------------------------------------------------------------------------------------------
val: True # (bool) validate/test during training
split: val # (str) dataset split to use for validation, i.e. 'val', 'test' or 'train'
save_json: False # (bool) save results to JSON file
save_hybrid: False # (bool) save hybrid version of labels (labels + additional predictions)
conf: # (float, optional) object confidence threshold for detection (default 0.25 predict, 0.001 val)
iou: 0.7 # (float) intersection over union (IoU) threshold for NMS
max_det: 300 # (int) maximum number of detections per image
half: False # (bool) use half precision (FP16)
dnn: False # (bool) use OpenCV DNN for ONNX inference
plots: True # (bool) save plots and images during train/val
# Predict settings -----------------------------------------------------------------------------------------------------
source: # (str, optional) source directory for images or videos
vid_stride: 1 # (int) video frame-rate stride
stream_buffer: False # (bool) buffer all streaming frames (True) or return the most recent frame (False)
visualize: False # (bool) visualize model features
augment: False # (bool) apply image augmentation to prediction sources
agnostic_nms: False # (bool) class-agnostic NMS
classes: # (int | list[int], optional) filter results by class, i.e. classes=0, or classes=[0,2,3]
retina_masks: False # (bool) use high-resolution segmentation masks
embed: # (list[int], optional) return feature vectors/embeddings from given layers
# Visualize settings ---------------------------------------------------------------------------------------------------
show: False # (bool) show predicted images and videos if environment allows
save_frames: False # (bool) save predicted individual video frames
save_txt: False # (bool) save results as .txt file
save_conf: False # (bool) save results with confidence scores
save_crop: False # (bool) save cropped images with results
show_labels: True # (bool) show prediction labels, i.e. 'person'
show_conf: True # (bool) show prediction confidence, i.e. '0.99'
show_boxes: True # (bool) show prediction boxes
line_width: # (int, optional) line width of the bounding boxes. Scaled to image size if None.
# Export settings ------------------------------------------------------------------------------------------------------
format: torchscript # (str) format to export to, choices at https://docs.ultralytics.com/modes/export/#export-formats
keras: False # (bool) use Kera=s
optimize: False # (bool) TorchScript: optimize for mobile
int8: False # (bool) CoreML/TF INT8 quantization
dynamic: False # (bool) ONNX/TF/TensorRT: dynamic axes
simplify: False # (bool) ONNX: simplify model using `onnxslim`
opset: # (int, optional) ONNX: opset version
workspace: 4 # (int) TensorRT: workspace size (GB)
nms: False # (bool) CoreML: add NMS
# Hyperparameters ------------------------------------------------------------------------------------------------------
lr0: 0.01 # (float) initial learning rate (i.e. SGD=1E-2, Adam=1E-3)
lrf: 0.01 # (float) final learning rate (lr0 * lrf)
momentum: 0.937 # (float) SGD momentum/Adam beta1
weight_decay: 0.0005 # (float) optimizer weight decay 5e-4
warmup_epochs: 3.0 # (float) warmup epochs (fractions ok)
warmup_momentum: 0.8 # (float) warmup initial momentum
warmup_bias_lr: 0.1 # (float) warmup initial bias lr
box: 7.5 # (float) box loss gain
cls: 0.5 # (float) cls loss gain (scale with pixels)
dfl: 1.5 # (float) dfl loss gain
pose: 12.0 # (float) pose loss gain
kobj: 1.0 # (float) keypoint obj loss gain
label_smoothing: 0.0 # (float) label smoothing (fraction)
nbs: 64 # (int) nominal batch size
hsv_h: 0.015 # (float) image HSV-Hue augmentation (fraction)
hsv_s: 0.7 # (float) image HSV-Saturation augmentation (fraction)
hsv_v: 0.4 # (float) image HSV-Value augmentation (fraction)
degrees: 0.0 # (float) image rotation (+/- deg)
translate: 0.1 # (float) image translation (+/- fraction)
scale: 0.5 # (float) image scale (+/- gain)
shear: 0.0 # (float) image shear (+/- deg)
perspective: 0.0 # (float) image perspective (+/- fraction), range 0-0.001
flipud: 0.0 # (float) image flip up-down (probability)
fliplr: 0.5 # (float) image flip left-right (probability)
bgr: 0.0 # (float) image channel BGR (probability)
mosaic: 1.0 # (float) image mosaic (probability)
mixup: 0.0 # (float) image mixup (probability)
copy_paste: 0.0 # (float) segment copy-paste (probability)
auto_augment: randaugment # (str) auto augmentation policy for classification (randaugment, autoaugment, augmix)
erasing: 0.4 # (float) probability of random erasing during classification training (0-1)
crop_fraction: 1.0 # (float) image crop fraction for classification evaluation/inference (0-1)
# Custom config.yaml ---------------------------------------------------------------------------------------------------
cfg: # (str, optional) for overriding defaults.yaml
# Tracker settings ------------------------------------------------------------------------------------------------------
tracker: botsort.yaml # (str) tracker type, choices=[botsort.yaml, bytetrack.yaml]

40
ultralytics/cfg/models/README.md
View File

@ -1,40 +0,0 @@
## Models
Welcome to the Ultralytics Models directory! Here you will find a wide variety of pre-configured model configuration files (`*.yaml`s) that can be used to create custom YOLO models. The models in this directory have been expertly crafted and fine-tuned by the Ultralytics team to provide the best performance for a wide range of object detection and image segmentation tasks.
These model configurations cover a wide range of scenarios, from simple object detection to more complex tasks like instance segmentation and object tracking. They are also designed to run efficiently on a variety of hardware platforms, from CPUs to GPUs. Whether you are a seasoned machine learning practitioner or just getting started with YOLO, this directory provides a great starting point for your custom model development needs.
To get started, simply browse through the models in this directory and find one that best suits your needs. Once you've selected a model, you can use the provided `*.yaml` file to train and deploy your custom YOLO model with ease. See full details at the Ultralytics [Docs](https://docs.ultralytics.com/models), and if you need help or have any questions, feel free to reach out to the Ultralytics team for support. So, don't wait, start creating your custom YOLO model now!
### Usage
Model `*.yaml` files may be used directly in the Command Line Interface (CLI) with a `yolo` command:
```bash
yolo task=detect mode=train model=yolov8n.yaml data=coco128.yaml epochs=100
```
They may also be used directly in a Python environment, and accepts the same [arguments](https://docs.ultralytics.com/usage/cfg/) as in the CLI example above:
```python
from ultralytics import YOLO
model = YOLO("model.yaml") # build a YOLOv8n model from scratch
# YOLO("model.pt") use pre-trained model if available
model.info() # display model information
model.train(data="coco128.yaml", epochs=100) # train the model
```
## Pre-trained Model Architectures
Ultralytics supports many model architectures. Visit https://docs.ultralytics.com/models to view detailed information and usage. Any of these models can be used by loading their configs or pretrained checkpoints if available.
## Contribute New Models
Have you trained a new YOLO variant or achieved state-of-the-art performance with specific tuning? We'd love to showcase your work in our Models section! Contributions from the community in the form of new models, architectures, or optimizations are highly valued and can significantly enrich our repository.
By contributing to this section, you're helping us offer a wider array of model choices and configurations to the community. It's a fantastic way to share your knowledge and expertise while making the Ultralytics YOLO ecosystem even more versatile.
To get started, please consult our [Contributing Guide](https://docs.ultralytics.com/help/contributing) for step-by-step instructions on how to submit a Pull Request (PR) 🛠️. Your contributions are eagerly awaited!
Let's join hands to extend the range and capabilities of the Ultralytics YOLO models 🙏!

50
ultralytics/cfg/models/rt-detr/rtdetr-l.yaml
View File

@ -1,50 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# RT-DETR-l object detection model with P3-P5 outputs. For details see https://docs.ultralytics.com/models/rtdetr
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n-cls.yaml' will call yolov8-cls.yaml with scale 'n'
# [depth, width, max_channels]
l: [1.00, 1.00, 1024]
backbone:
# [from, repeats, module, args]
- [-1, 1, HGStem, [32, 48]] # 0-P2/4
- [-1, 6, HGBlock, [48, 128, 3]] # stage 1
- [-1, 1, DWConv, [128, 3, 2, 1, False]] # 2-P3/8
- [-1, 6, HGBlock, [96, 512, 3]] # stage 2
- [-1, 1, DWConv, [512, 3, 2, 1, False]] # 4-P3/16
- [-1, 6, HGBlock, [192, 1024, 5, True, False]] # cm, c2, k, light, shortcut
- [-1, 6, HGBlock, [192, 1024, 5, True, True]]
- [-1, 6, HGBlock, [192, 1024, 5, True, True]] # stage 3
- [-1, 1, DWConv, [1024, 3, 2, 1, False]] # 8-P4/32
- [-1, 6, HGBlock, [384, 2048, 5, True, False]] # stage 4
head:
- [-1, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 10 input_proj.2
- [-1, 1, AIFI, [1024, 8]]
- [-1, 1, Conv, [256, 1, 1]] # 12, Y5, lateral_convs.0
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [7, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 14 input_proj.1
- [[-2, -1], 1, Concat, [1]]
- [-1, 3, RepC3, [256]] # 16, fpn_blocks.0
- [-1, 1, Conv, [256, 1, 1]] # 17, Y4, lateral_convs.1
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [3, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 19 input_proj.0
- [[-2, -1], 1, Concat, [1]] # cat backbone P4
- [-1, 3, RepC3, [256]] # X3 (21), fpn_blocks.1
- [-1, 1, Conv, [256, 3, 2]] # 22, downsample_convs.0
- [[-1, 17], 1, Concat, [1]] # cat Y4
- [-1, 3, RepC3, [256]] # F4 (24), pan_blocks.0
- [-1, 1, Conv, [256, 3, 2]] # 25, downsample_convs.1
- [[-1, 12], 1, Concat, [1]] # cat Y5
- [-1, 3, RepC3, [256]] # F5 (27), pan_blocks.1
- [[21, 24, 27], 1, RTDETRDecoder, [nc]] # Detect(P3, P4, P5)

42
ultralytics/cfg/models/rt-detr/rtdetr-resnet101.yaml
View File

@ -1,42 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# RT-DETR-ResNet101 object detection model with P3-P5 outputs.
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n-cls.yaml' will call yolov8-cls.yaml with scale 'n'
# [depth, width, max_channels]
l: [1.00, 1.00, 1024]
backbone:
# [from, repeats, module, args]
- [-1, 1, ResNetLayer, [3, 64, 1, True, 1]] # 0
- [-1, 1, ResNetLayer, [64, 64, 1, False, 3]] # 1
- [-1, 1, ResNetLayer, [256, 128, 2, False, 4]] # 2
- [-1, 1, ResNetLayer, [512, 256, 2, False, 23]] # 3
- [-1, 1, ResNetLayer, [1024, 512, 2, False, 3]] # 4
head:
- [-1, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 5
- [-1, 1, AIFI, [1024, 8]]
- [-1, 1, Conv, [256, 1, 1]] # 7
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [3, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 9
- [[-2, -1], 1, Concat, [1]]
- [-1, 3, RepC3, [256]] # 11
- [-1, 1, Conv, [256, 1, 1]] # 12
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [2, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 14
- [[-2, -1], 1, Concat, [1]] # cat backbone P4
- [-1, 3, RepC3, [256]] # X3 (16), fpn_blocks.1
- [-1, 1, Conv, [256, 3, 2]] # 17, downsample_convs.0
- [[-1, 12], 1, Concat, [1]] # cat Y4
- [-1, 3, RepC3, [256]] # F4 (19), pan_blocks.0
- [-1, 1, Conv, [256, 3, 2]] # 20, downsample_convs.1
- [[-1, 7], 1, Concat, [1]] # cat Y5
- [-1, 3, RepC3, [256]] # F5 (22), pan_blocks.1
- [[16, 19, 22], 1, RTDETRDecoder, [nc]] # Detect(P3, P4, P5)

42
ultralytics/cfg/models/rt-detr/rtdetr-resnet50.yaml
View File

@ -1,42 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# RT-DETR-ResNet50 object detection model with P3-P5 outputs.
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n-cls.yaml' will call yolov8-cls.yaml with scale 'n'
# [depth, width, max_channels]
l: [1.00, 1.00, 1024]
backbone:
# [from, repeats, module, args]
- [-1, 1, ResNetLayer, [3, 64, 1, True, 1]] # 0
- [-1, 1, ResNetLayer, [64, 64, 1, False, 3]] # 1
- [-1, 1, ResNetLayer, [256, 128, 2, False, 4]] # 2
- [-1, 1, ResNetLayer, [512, 256, 2, False, 6]] # 3
- [-1, 1, ResNetLayer, [1024, 512, 2, False, 3]] # 4
head:
- [-1, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 5
- [-1, 1, AIFI, [1024, 8]]
- [-1, 1, Conv, [256, 1, 1]] # 7
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [3, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 9
- [[-2, -1], 1, Concat, [1]]
- [-1, 3, RepC3, [256]] # 11
- [-1, 1, Conv, [256, 1, 1]] # 12
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [2, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 14
- [[-2, -1], 1, Concat, [1]] # cat backbone P4
- [-1, 3, RepC3, [256]] # X3 (16), fpn_blocks.1
- [-1, 1, Conv, [256, 3, 2]] # 17, downsample_convs.0
- [[-1, 12], 1, Concat, [1]] # cat Y4
- [-1, 3, RepC3, [256]] # F4 (19), pan_blocks.0
- [-1, 1, Conv, [256, 3, 2]] # 20, downsample_convs.1
- [[-1, 7], 1, Concat, [1]] # cat Y5
- [-1, 3, RepC3, [256]] # F5 (22), pan_blocks.1
- [[16, 19, 22], 1, RTDETRDecoder, [nc]] # Detect(P3, P4, P5)

54
ultralytics/cfg/models/rt-detr/rtdetr-x.yaml
View File

@ -1,54 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# RT-DETR-x object detection model with P3-P5 outputs. For details see https://docs.ultralytics.com/models/rtdetr
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n-cls.yaml' will call yolov8-cls.yaml with scale 'n'
# [depth, width, max_channels]
x: [1.00, 1.00, 2048]
backbone:
# [from, repeats, module, args]
- [-1, 1, HGStem, [32, 64]] # 0-P2/4
- [-1, 6, HGBlock, [64, 128, 3]] # stage 1
- [-1, 1, DWConv, [128, 3, 2, 1, False]] # 2-P3/8
- [-1, 6, HGBlock, [128, 512, 3]]
- [-1, 6, HGBlock, [128, 512, 3, False, True]] # 4-stage 2
- [-1, 1, DWConv, [512, 3, 2, 1, False]] # 5-P3/16
- [-1, 6, HGBlock, [256, 1024, 5, True, False]] # cm, c2, k, light, shortcut
- [-1, 6, HGBlock, [256, 1024, 5, True, True]]
- [-1, 6, HGBlock, [256, 1024, 5, True, True]]
- [-1, 6, HGBlock, [256, 1024, 5, True, True]]
- [-1, 6, HGBlock, [256, 1024, 5, True, True]] # 10-stage 3
- [-1, 1, DWConv, [1024, 3, 2, 1, False]] # 11-P4/32
- [-1, 6, HGBlock, [512, 2048, 5, True, False]]
- [-1, 6, HGBlock, [512, 2048, 5, True, True]] # 13-stage 4
head:
- [-1, 1, Conv, [384, 1, 1, None, 1, 1, False]] # 14 input_proj.2
- [-1, 1, AIFI, [2048, 8]]
- [-1, 1, Conv, [384, 1, 1]] # 16, Y5, lateral_convs.0
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [10, 1, Conv, [384, 1, 1, None, 1, 1, False]] # 18 input_proj.1
- [[-2, -1], 1, Concat, [1]]
- [-1, 3, RepC3, [384]] # 20, fpn_blocks.0
- [-1, 1, Conv, [384, 1, 1]] # 21, Y4, lateral_convs.1
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [4, 1, Conv, [384, 1, 1, None, 1, 1, False]] # 23 input_proj.0
- [[-2, -1], 1, Concat, [1]] # cat backbone P4
- [-1, 3, RepC3, [384]] # X3 (25), fpn_blocks.1
- [-1, 1, Conv, [384, 3, 2]] # 26, downsample_convs.0
- [[-1, 21], 1, Concat, [1]] # cat Y4
- [-1, 3, RepC3, [384]] # F4 (28), pan_blocks.0
- [-1, 1, Conv, [384, 3, 2]] # 29, downsample_convs.1
- [[-1, 16], 1, Concat, [1]] # cat Y5
- [-1, 3, RepC3, [384]] # F5 (31), pan_blocks.1
- [[25, 28, 31], 1, RTDETRDecoder, [nc]] # Detect(P3, P4, P5)

40
ultralytics/cfg/models/v10/yolov10b-seg.yaml
View File

@ -1,40 +0,0 @@
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
b: [0.67, 1.00, 512]
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, SCDown, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, SCDown, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2fCIB, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
- [-1, 1, PSA, [1024]] # 10
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2fCIB, [512, True]] # 13
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2f, [256]] # 16 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 13], 1, Concat, [1]] # cat head P4
- [-1, 3, C2fCIB, [512, True]] # 19 (P4/16-medium)
- [-1, 1, SCDown, [512, 3, 2]]
- [[-1, 10], 1, Concat, [1]] # cat head P5
- [-1, 3, C2fCIB, [1024, True]] # 22 (P5/32-large)
- [[16, 19, 22], 1, v10Segment, [nc, 32, 256]] # segment(P3, P4, P5)

40
ultralytics/cfg/models/v10/yolov10b.yaml
View File

@ -1,40 +0,0 @@
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
b: [0.67, 1.00, 512]
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, SCDown, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, SCDown, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2fCIB, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
- [-1, 1, PSA, [1024]] # 10
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2fCIB, [512, True]] # 13
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2f, [256]] # 16 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 13], 1, Concat, [1]] # cat head P4
- [-1, 3, C2fCIB, [512, True]] # 19 (P4/16-medium)
- [-1, 1, SCDown, [512, 3, 2]]
- [[-1, 10], 1, Concat, [1]] # cat head P5
- [-1, 3, C2fCIB, [1024, True]] # 22 (P5/32-large)
- [[16, 19, 22], 1, v10Detect, [nc]] # Detect(P3, P4, P5)

40
ultralytics/cfg/models/v10/yolov10l-seg.yaml
View File

@ -1,40 +0,0 @@
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
l: [1.00, 1.00, 512] # YOLOv8l summary: 365 layers, 43691520 parameters, 43691504 gradients, 165.7 GFLOPs
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, SCDown, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, SCDown, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2fCIB, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
- [-1, 1, PSA, [1024]] # 10
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2fCIB, [512, True]] # 13
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2f, [256]] # 16 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 13], 1, Concat, [1]] # cat head P4
- [-1, 3, C2fCIB, [512, True]] # 19 (P4/16-medium)
- [-1, 1, SCDown, [512, 3, 2]]
- [[-1, 10], 1, Concat, [1]] # cat head P5
- [-1, 3, C2fCIB, [1024, True]] # 22 (P5/32-large)
- [[16, 19, 22], 1, v10Segment, [nc, 32, 256]] # segment(P3, P4, P5)

40
ultralytics/cfg/models/v10/yolov10l.yaml
View File

@ -1,40 +0,0 @@
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
l: [1.00, 1.00, 512] # YOLOv8l summary: 365 layers, 43691520 parameters, 43691504 gradients, 165.7 GFLOPs
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, SCDown, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, SCDown, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2fCIB, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
- [-1, 1, PSA, [1024]] # 10
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2fCIB, [512, True]] # 13
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2f, [256]] # 16 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 13], 1, Concat, [1]] # cat head P4
- [-1, 3, C2fCIB, [512, True]] # 19 (P4/16-medium)
- [-1, 1, SCDown, [512, 3, 2]]
- [[-1, 10], 1, Concat, [1]] # cat head P5
- [-1, 3, C2fCIB, [1024, True]] # 22 (P5/32-large)
- [[16, 19, 22], 1, v10Detect, [nc]] # Detect(P3, P4, P5)

43
ultralytics/cfg/models/v10/yolov10m-seg.yaml
View File

@ -1,43 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8 object detection model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
m: [0.67, 0.75, 768] # YOLOv8m summary: 295 layers, 25902640 parameters, 25902624 gradients, 79.3 GFLOPs
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, SCDown, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, SCDown, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2fCIB, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
- [-1, 1, PSA, [1024]] # 10
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2f, [512]] # 13
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2f, [256]] # 16 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 13], 1, Concat, [1]] # cat head P4
- [-1, 3, C2fCIB, [512, True]] # 19 (P4/16-medium)
- [-1, 1, SCDown, [512, 3, 2]]
- [[-1, 10], 1, Concat, [1]] # cat head P5
- [-1, 3, C2fCIB, [1024, True]] # 22 (P5/32-large)
- [[16, 19, 22], 1, v10Segment, [nc, 32, 256]] # segment(P3, P4, P5)

43
ultralytics/cfg/models/v10/yolov10m.yaml
View File

@ -1,43 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8 object detection model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
m: [0.67, 0.75, 768] # YOLOv8m summary: 295 layers, 25902640 parameters, 25902624 gradients, 79.3 GFLOPs
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, SCDown, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, SCDown, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2fCIB, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
- [-1, 1, PSA, [1024]] # 10
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2f, [512]] # 13
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2f, [256]] # 16 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 13], 1, Concat, [1]] # cat head P4
- [-1, 3, C2fCIB, [512, True]] # 19 (P4/16-medium)
- [-1, 1, SCDown, [512, 3, 2]]
- [[-1, 10], 1, Concat, [1]] # cat head P5
- [-1, 3, C2fCIB, [1024, True]] # 22 (P5/32-large)
- [[16, 19, 22], 1, v10Detect, [nc]] # Detect(P3, P4, P5)

40
ultralytics/cfg/models/v10/yolov10n-seg.yaml
View File

@ -1,40 +0,0 @@
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024]
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, SCDown, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, SCDown, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2f, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
- [-1, 1, PSA, [1024]] # 10
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2f, [512]] # 13
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2f, [256]] # 16 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 13], 1, Concat, [1]] # cat head P4
- [-1, 3, C2f, [512]] # 19 (P4/16-medium)
- [-1, 1, SCDown, [512, 3, 2]]
- [[-1, 10], 1, Concat, [1]] # cat head P5
- [-1, 3, C2fCIB, [1024, True, True]] # 22 (P5/32-large)
- [[16, 19, 22], 1, v10Segment, [nc, 32, 256]] # segment(P3, P4, P5)

40
ultralytics/cfg/models/v10/yolov10n.yaml
View File

@ -1,40 +0,0 @@
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024]
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, SCDown, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, SCDown, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2f, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
- [-1, 1, PSA, [1024]] # 10
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2f, [512]] # 13
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2f, [256]] # 16 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 13], 1, Concat, [1]] # cat head P4
- [-1, 3, C2f, [512]] # 19 (P4/16-medium)
- [-1, 1, SCDown, [512, 3, 2]]
- [[-1, 10], 1, Concat, [1]] # cat head P5
- [-1, 3, C2fCIB, [1024, True, True]] # 22 (P5/32-large)
- [[16, 19, 22], 1, v10Detect, [nc]] # Detect(P3, P4, P5)

39
ultralytics/cfg/models/v10/yolov10s-seg.yaml
View File

@ -1,39 +0,0 @@
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
s: [0.33, 0.50, 1024]
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, SCDown, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, SCDown, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2fCIB, [1024, True, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
- [-1, 1, PSA, [1024]] # 10
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2f, [512]] # 13
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2f, [256]] # 16 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 13], 1, Concat, [1]] # cat head P4
- [-1, 3, C2f, [512]] # 19 (P4/16-medium)
- [-1, 1, SCDown, [512, 3, 2]]
- [[-1, 10], 1, Concat, [1]] # cat head P5
- [-1, 3, C2fCIB, [1024, True, True]] # 22 (P5/32-large)
- [[16, 19, 22], 1, v10Segment, [nc, 32, 256]] # segment(P3, P4, P5)

39
ultralytics/cfg/models/v10/yolov10s.yaml
View File

@ -1,39 +0,0 @@
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
s: [0.33, 0.50, 1024]
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, SCDown, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, SCDown, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2fCIB, [1024, True, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
- [-1, 1, PSA, [1024]] # 10
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2f, [512]] # 13
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2f, [256]] # 16 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 13], 1, Concat, [1]] # cat head P4
- [-1, 3, C2f, [512]] # 19 (P4/16-medium)
- [-1, 1, SCDown, [512, 3, 2]]
- [[-1, 10], 1, Concat, [1]] # cat head P5
- [-1, 3, C2fCIB, [1024, True, True]] # 22 (P5/32-large)
- [[16, 19, 22], 1, v10Detect, [nc]] # Detect(P3, P4, P5)

40
ultralytics/cfg/models/v10/yolov10x-seg.yaml
View File

@ -1,40 +0,0 @@
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
x: [1.00, 1.25, 512]
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, SCDown, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2fCIB, [512, True]]
- [-1, 1, SCDown, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2fCIB, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
- [-1, 1, PSA, [1024]] # 10
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2fCIB, [512, True]] # 13
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2f, [256]] # 16 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 13], 1, Concat, [1]] # cat head P4
- [-1, 3, C2fCIB, [512, True]] # 19 (P4/16-medium)
- [-1, 1, SCDown, [512, 3, 2]]
- [[-1, 10], 1, Concat, [1]] # cat head P5
- [-1, 3, C2fCIB, [1024, True]] # 22 (P5/32-large)
- [[16, 19, 22], 1, v10Segment, [nc, 32, 256]] # segment(P3, P4, P5)

40
ultralytics/cfg/models/v10/yolov10x.yaml
View File

@ -1,40 +0,0 @@
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
x: [1.00, 1.25, 512]
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, SCDown, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2fCIB, [512, True]]
- [-1, 1, SCDown, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2fCIB, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
- [-1, 1, PSA, [1024]] # 10
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2fCIB, [512, True]] # 13
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2f, [256]] # 16 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 13], 1, Concat, [1]] # cat head P4
- [-1, 3, C2fCIB, [512, True]] # 19 (P4/16-medium)
- [-1, 1, SCDown, [512, 3, 2]]
- [[-1, 10], 1, Concat, [1]] # cat head P5
- [-1, 3, C2fCIB, [1024, True]] # 22 (P5/32-large)
- [[16, 19, 22], 1, v10Detect, [nc]] # Detect(P3, P4, P5)

46
ultralytics/cfg/models/v3/yolov3-spp.yaml
View File

@ -1,46 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv3-SPP object detection model with P3-P5 outputs. For details see https://docs.ultralytics.com/models/yolov3
# Parameters
nc: 80 # number of classes
depth_multiple: 1.0 # model depth multiple
width_multiple: 1.0 # layer channel multiple
# darknet53 backbone
backbone:
# [from, number, module, args]
- [-1, 1, Conv, [32, 3, 1]] # 0
- [-1, 1, Conv, [64, 3, 2]] # 1-P1/2
- [-1, 1, Bottleneck, [64]]
- [-1, 1, Conv, [128, 3, 2]] # 3-P2/4
- [-1, 2, Bottleneck, [128]]
- [-1, 1, Conv, [256, 3, 2]] # 5-P3/8
- [-1, 8, Bottleneck, [256]]
- [-1, 1, Conv, [512, 3, 2]] # 7-P4/16
- [-1, 8, Bottleneck, [512]]
- [-1, 1, Conv, [1024, 3, 2]] # 9-P5/32
- [-1, 4, Bottleneck, [1024]] # 10
# YOLOv3-SPP head
head:
- [-1, 1, Bottleneck, [1024, False]]
- [-1, 1, SPP, [512, [5, 9, 13]]]
- [-1, 1, Conv, [1024, 3, 1]]
- [-1, 1, Conv, [512, 1, 1]]
- [-1, 1, Conv, [1024, 3, 1]] # 15 (P5/32-large)
- [-2, 1, Conv, [256, 1, 1]]
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 8], 1, Concat, [1]] # cat backbone P4
- [-1, 1, Bottleneck, [512, False]]
- [-1, 1, Bottleneck, [512, False]]
- [-1, 1, Conv, [256, 1, 1]]
- [-1, 1, Conv, [512, 3, 1]] # 22 (P4/16-medium)
- [-2, 1, Conv, [128, 1, 1]]
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P3
- [-1, 1, Bottleneck, [256, False]]
- [-1, 2, Bottleneck, [256, False]] # 27 (P3/8-small)
- [[27, 22, 15], 1, Detect, [nc]] # Detect(P3, P4, P5)

37
ultralytics/cfg/models/v3/yolov3-tiny.yaml
View File

@ -1,37 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv3-tiny object detection model with P4-P5 outputs. For details see https://docs.ultralytics.com/models/yolov3
# Parameters
nc: 80 # number of classes
depth_multiple: 1.0 # model depth multiple
width_multiple: 1.0 # layer channel multiple
# YOLOv3-tiny backbone
backbone:
# [from, number, module, args]
- [-1, 1, Conv, [16, 3, 1]] # 0
- [-1, 1, nn.MaxPool2d, [2, 2, 0]] # 1-P1/2
- [-1, 1, Conv, [32, 3, 1]]
- [-1, 1, nn.MaxPool2d, [2, 2, 0]] # 3-P2/4
- [-1, 1, Conv, [64, 3, 1]]
- [-1, 1, nn.MaxPool2d, [2, 2, 0]] # 5-P3/8
- [-1, 1, Conv, [128, 3, 1]]
- [-1, 1, nn.MaxPool2d, [2, 2, 0]] # 7-P4/16
- [-1, 1, Conv, [256, 3, 1]]
- [-1, 1, nn.MaxPool2d, [2, 2, 0]] # 9-P5/32
- [-1, 1, Conv, [512, 3, 1]]
- [-1, 1, nn.ZeroPad2d, [[0, 1, 0, 1]]] # 11
- [-1, 1, nn.MaxPool2d, [2, 1, 0]] # 12
# YOLOv3-tiny head
head:
- [-1, 1, Conv, [1024, 3, 1]]
- [-1, 1, Conv, [256, 1, 1]]
- [-1, 1, Conv, [512, 3, 1]] # 15 (P5/32-large)
- [-2, 1, Conv, [128, 1, 1]]
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 8], 1, Concat, [1]] # cat backbone P4
- [-1, 1, Conv, [256, 3, 1]] # 19 (P4/16-medium)
- [[19, 15], 1, Detect, [nc]] # Detect(P4, P5)

46
ultralytics/cfg/models/v3/yolov3.yaml
View File

@ -1,46 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv3 object detection model with P3-P5 outputs. For details see https://docs.ultralytics.com/models/yolov3
# Parameters
nc: 80 # number of classes
depth_multiple: 1.0 # model depth multiple
width_multiple: 1.0 # layer channel multiple
# darknet53 backbone
backbone:
# [from, number, module, args]
- [-1, 1, Conv, [32, 3, 1]] # 0
- [-1, 1, Conv, [64, 3, 2]] # 1-P1/2
- [-1, 1, Bottleneck, [64]]
- [-1, 1, Conv, [128, 3, 2]] # 3-P2/4
- [-1, 2, Bottleneck, [128]]
- [-1, 1, Conv, [256, 3, 2]] # 5-P3/8
- [-1, 8, Bottleneck, [256]]
- [-1, 1, Conv, [512, 3, 2]] # 7-P4/16
- [-1, 8, Bottleneck, [512]]
- [-1, 1, Conv, [1024, 3, 2]] # 9-P5/32
- [-1, 4, Bottleneck, [1024]] # 10
# YOLOv3 head
head:
- [-1, 1, Bottleneck, [1024, False]]
- [-1, 1, Conv, [512, 1, 1]]
- [-1, 1, Conv, [1024, 3, 1]]
- [-1, 1, Conv, [512, 1, 1]]
- [-1, 1, Conv, [1024, 3, 1]] # 15 (P5/32-large)
- [-2, 1, Conv, [256, 1, 1]]
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 8], 1, Concat, [1]] # cat backbone P4
- [-1, 1, Bottleneck, [512, False]]
- [-1, 1, Bottleneck, [512, False]]
- [-1, 1, Conv, [256, 1, 1]]
- [-1, 1, Conv, [512, 3, 1]] # 22 (P4/16-medium)
- [-2, 1, Conv, [128, 1, 1]]
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P3
- [-1, 1, Bottleneck, [256, False]]
- [-1, 2, Bottleneck, [256, False]] # 27 (P3/8-small)
- [[27, 22, 15], 1, Detect, [nc]] # Detect(P3, P4, P5)

59
ultralytics/cfg/models/v5/yolov5-p6.yaml
View File

@ -1,59 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv5 object detection model with P3-P6 outputs. For details see https://docs.ultralytics.com/models/yolov5
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov5n-p6.yaml' will call yolov5-p6.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024]
s: [0.33, 0.50, 1024]
m: [0.67, 0.75, 1024]
l: [1.00, 1.00, 1024]
x: [1.33, 1.25, 1024]
# YOLOv5 v6.0 backbone
backbone:
# [from, number, module, args]
- [-1, 1, Conv, [64, 6, 2, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C3, [128]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C3, [256]]
- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
- [-1, 9, C3, [512]]
- [-1, 1, Conv, [768, 3, 2]] # 7-P5/32
- [-1, 3, C3, [768]]
- [-1, 1, Conv, [1024, 3, 2]] # 9-P6/64
- [-1, 3, C3, [1024]]
- [-1, 1, SPPF, [1024, 5]] # 11
# YOLOv5 v6.0 head
head:
- [-1, 1, Conv, [768, 1, 1]]
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 8], 1, Concat, [1]] # cat backbone P5
- [-1, 3, C3, [768, False]] # 15
- [-1, 1, Conv, [512, 1, 1]]
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C3, [512, False]] # 19
- [-1, 1, Conv, [256, 1, 1]]
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C3, [256, False]] # 23 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 20], 1, Concat, [1]] # cat head P4
- [-1, 3, C3, [512, False]] # 26 (P4/16-medium)
- [-1, 1, Conv, [512, 3, 2]]
- [[-1, 16], 1, Concat, [1]] # cat head P5
- [-1, 3, C3, [768, False]] # 29 (P5/32-large)
- [-1, 1, Conv, [768, 3, 2]]
- [[-1, 12], 1, Concat, [1]] # cat head P6
- [-1, 3, C3, [1024, False]] # 32 (P6/64-xlarge)
- [[23, 26, 29, 32], 1, Detect, [nc]] # Detect(P3, P4, P5, P6)

48
ultralytics/cfg/models/v5/yolov5.yaml
View File

@ -1,48 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv5 object detection model with P3-P5 outputs. For details see https://docs.ultralytics.com/models/yolov5
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov5n.yaml' will call yolov5.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024]
s: [0.33, 0.50, 1024]
m: [0.67, 0.75, 1024]
l: [1.00, 1.00, 1024]
x: [1.33, 1.25, 1024]
# YOLOv5 v6.0 backbone
backbone:
# [from, number, module, args]
- [-1, 1, Conv, [64, 6, 2, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C3, [128]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C3, [256]]
- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
- [-1, 9, C3, [512]]
- [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C3, [1024]]
- [-1, 1, SPPF, [1024, 5]] # 9
# YOLOv5 v6.0 head
head:
- [-1, 1, Conv, [512, 1, 1]]
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C3, [512, False]] # 13
- [-1, 1, Conv, [256, 1, 1]]
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C3, [256, False]] # 17 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 14], 1, Concat, [1]] # cat head P4
- [-1, 3, C3, [512, False]] # 20 (P4/16-medium)
- [-1, 1, Conv, [512, 3, 2]]
- [[-1, 10], 1, Concat, [1]] # cat head P5
- [-1, 3, C3, [1024, False]] # 23 (P5/32-large)
- [[17, 20, 23], 1, Detect, [nc]] # Detect(P3, P4, P5)

53
ultralytics/cfg/models/v6/yolov6.yaml
View File

@ -1,53 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv6 object detection model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/models/yolov6
# Parameters
nc: 80 # number of classes
activation: nn.ReLU() # (optional) model default activation function
scales: # model compound scaling constants, i.e. 'model=yolov6n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024]
s: [0.33, 0.50, 1024]
m: [0.67, 0.75, 768]
l: [1.00, 1.00, 512]
x: [1.00, 1.25, 512]
# YOLOv6-3.0s backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 6, Conv, [128, 3, 1]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 12, Conv, [256, 3, 1]]
- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
- [-1, 18, Conv, [512, 3, 1]]
- [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32
- [-1, 6, Conv, [1024, 3, 1]]
- [-1, 1, SPPF, [1024, 5]] # 9
# YOLOv6-3.0s head
head:
- [-1, 1, Conv, [256, 1, 1]]
- [-1, 1, nn.ConvTranspose2d, [256, 2, 2, 0]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 1, Conv, [256, 3, 1]]
- [-1, 9, Conv, [256, 3, 1]] # 14
- [-1, 1, Conv, [128, 1, 1]]
- [-1, 1, nn.ConvTranspose2d, [128, 2, 2, 0]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 1, Conv, [128, 3, 1]]
- [-1, 9, Conv, [128, 3, 1]] # 19
- [-1, 1, Conv, [128, 3, 2]]
- [[-1, 15], 1, Concat, [1]] # cat head P4
- [-1, 1, Conv, [256, 3, 1]]
- [-1, 9, Conv, [256, 3, 1]] # 23
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 10], 1, Concat, [1]] # cat head P5
- [-1, 1, Conv, [512, 3, 1]]
- [-1, 9, Conv, [512, 3, 1]] # 27
- [[19, 23, 27], 1, Detect, [nc]] # Detect(P3, P4, P5)

25
ultralytics/cfg/models/v8/yolov8-cls-resnet101.yaml
View File

@ -1,25 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8-cls image classification model. For Usage examples see https://docs.ultralytics.com/tasks/classify
# Parameters
nc: 1000 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n-cls.yaml' will call yolov8-cls.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024]
s: [0.33, 0.50, 1024]
m: [0.67, 0.75, 1024]
l: [1.00, 1.00, 1024]
x: [1.00, 1.25, 1024]
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, ResNetLayer, [3, 64, 1, True, 1]] # 0-P1/2
- [-1, 1, ResNetLayer, [64, 64, 1, False, 3]] # 1-P2/4
- [-1, 1, ResNetLayer, [256, 128, 2, False, 4]] # 2-P3/8
- [-1, 1, ResNetLayer, [512, 256, 2, False, 23]] # 3-P4/16
- [-1, 1, ResNetLayer, [1024, 512, 2, False, 3]] # 4-P5/32
# YOLOv8.0n head
head:
- [-1, 1, Classify, [nc]] # Classify

25
ultralytics/cfg/models/v8/yolov8-cls-resnet50.yaml
View File

@ -1,25 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8-cls image classification model. For Usage examples see https://docs.ultralytics.com/tasks/classify
# Parameters
nc: 1000 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n-cls.yaml' will call yolov8-cls.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024]
s: [0.33, 0.50, 1024]
m: [0.67, 0.75, 1024]
l: [1.00, 1.00, 1024]
x: [1.00, 1.25, 1024]
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, ResNetLayer, [3, 64, 1, True, 1]] # 0-P1/2
- [-1, 1, ResNetLayer, [64, 64, 1, False, 3]] # 1-P2/4
- [-1, 1, ResNetLayer, [256, 128, 2, False, 4]] # 2-P3/8
- [-1, 1, ResNetLayer, [512, 256, 2, False, 6]] # 3-P4/16
- [-1, 1, ResNetLayer, [1024, 512, 2, False, 3]] # 4-P5/32
# YOLOv8.0n head
head:
- [-1, 1, Classify, [nc]] # Classify

29
ultralytics/cfg/models/v8/yolov8-cls.yaml
View File

@ -1,29 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8-cls image classification model. For Usage examples see https://docs.ultralytics.com/tasks/classify
# Parameters
nc: 1000 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n-cls.yaml' will call yolov8-cls.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024]
s: [0.33, 0.50, 1024]
m: [0.67, 0.75, 1024]
l: [1.00, 1.00, 1024]
x: [1.00, 1.25, 1024]
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2f, [1024, True]]
# YOLOv8.0n head
head:
- [-1, 1, Classify, [nc]] # Classify

54
ultralytics/cfg/models/v8/yolov8-ghost-p2.yaml
View File

@ -1,54 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8 object detection model with P2-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024] # YOLOv8n-ghost-p2 summary: 491 layers, 2033944 parameters, 2033928 gradients, 13.8 GFLOPs
s: [0.33, 0.50, 1024] # YOLOv8s-ghost-p2 summary: 491 layers, 5562080 parameters, 5562064 gradients, 25.1 GFLOPs
m: [0.67, 0.75, 768] # YOLOv8m-ghost-p2 summary: 731 layers, 9031728 parameters, 9031712 gradients, 42.8 GFLOPs
l: [1.00, 1.00, 512] # YOLOv8l-ghost-p2 summary: 971 layers, 12214448 parameters, 12214432 gradients, 69.1 GFLOPs
x: [1.00, 1.25, 512] # YOLOv8x-ghost-p2 summary: 971 layers, 18664776 parameters, 18664760 gradients, 103.3 GFLOPs
# YOLOv8.0-ghost backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, GhostConv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C3Ghost, [128, True]]
- [-1, 1, GhostConv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C3Ghost, [256, True]]
- [-1, 1, GhostConv, [512, 3, 2]] # 5-P4/16
- [-1, 6, C3Ghost, [512, True]]
- [-1, 1, GhostConv, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C3Ghost, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
# YOLOv8.0-ghost-p2 head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C3Ghost, [512]] # 12
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C3Ghost, [256]] # 15 (P3/8-small)
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 2], 1, Concat, [1]] # cat backbone P2
- [-1, 3, C3Ghost, [128]] # 18 (P2/4-xsmall)
- [-1, 1, GhostConv, [128, 3, 2]]
- [[-1, 15], 1, Concat, [1]] # cat head P3
- [-1, 3, C3Ghost, [256]] # 21 (P3/8-small)
- [-1, 1, GhostConv, [256, 3, 2]]
- [[-1, 12], 1, Concat, [1]] # cat head P4
- [-1, 3, C3Ghost, [512]] # 24 (P4/16-medium)
- [-1, 1, GhostConv, [512, 3, 2]]
- [[-1, 9], 1, Concat, [1]] # cat head P5
- [-1, 3, C3Ghost, [1024]] # 27 (P5/32-large)
- [[18, 21, 24, 27], 1, Detect, [nc]] # Detect(P2, P3, P4, P5)

56
ultralytics/cfg/models/v8/yolov8-ghost-p6.yaml
View File

@ -1,56 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8 object detection model with P3-P6 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n-p6.yaml' will call yolov8-p6.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024] # YOLOv8n-ghost-p6 summary: 529 layers, 2901100 parameters, 2901084 gradients, 5.8 GFLOPs
s: [0.33, 0.50, 1024] # YOLOv8s-ghost-p6 summary: 529 layers, 9520008 parameters, 9519992 gradients, 16.4 GFLOPs
m: [0.67, 0.75, 768] # YOLOv8m-ghost-p6 summary: 789 layers, 18002904 parameters, 18002888 gradients, 34.4 GFLOPs
l: [1.00, 1.00, 512] # YOLOv8l-ghost-p6 summary: 1049 layers, 21227584 parameters, 21227568 gradients, 55.3 GFLOPs
x: [1.00, 1.25, 512] # YOLOv8x-ghost-p6 summary: 1049 layers, 33057852 parameters, 33057836 gradients, 85.7 GFLOPs
# YOLOv8.0-ghost backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, GhostConv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C3Ghost, [128, True]]
- [-1, 1, GhostConv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C3Ghost, [256, True]]
- [-1, 1, GhostConv, [512, 3, 2]] # 5-P4/16
- [-1, 6, C3Ghost, [512, True]]
- [-1, 1, GhostConv, [768, 3, 2]] # 7-P5/32
- [-1, 3, C3Ghost, [768, True]]
- [-1, 1, GhostConv, [1024, 3, 2]] # 9-P6/64
- [-1, 3, C3Ghost, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 11
# YOLOv8.0-ghost-p6 head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 8], 1, Concat, [1]] # cat backbone P5
- [-1, 3, C3Ghost, [768]] # 14
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C3Ghost, [512]] # 17
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C3Ghost, [256]] # 20 (P3/8-small)
- [-1, 1, GhostConv, [256, 3, 2]]
- [[-1, 17], 1, Concat, [1]] # cat head P4
- [-1, 3, C3Ghost, [512]] # 23 (P4/16-medium)
- [-1, 1, GhostConv, [512, 3, 2]]
- [[-1, 14], 1, Concat, [1]] # cat head P5
- [-1, 3, C3Ghost, [768]] # 26 (P5/32-large)
- [-1, 1, GhostConv, [768, 3, 2]]
- [[-1, 11], 1, Concat, [1]] # cat head P6
- [-1, 3, C3Ghost, [1024]] # 29 (P6/64-xlarge)
- [[20, 23, 26, 29], 1, Detect, [nc]] # Detect(P3, P4, P5, P6)

47
ultralytics/cfg/models/v8/yolov8-ghost.yaml
View File

@ -1,47 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8 object detection model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
# Employs Ghost convolutions and modules proposed in Huawei's GhostNet in https://arxiv.org/abs/1911.11907v2
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024] # YOLOv8n-ghost summary: 403 layers, 1865316 parameters, 1865300 gradients, 5.8 GFLOPs
s: [0.33, 0.50, 1024] # YOLOv8s-ghost summary: 403 layers, 5960072 parameters, 5960056 gradients, 16.4 GFLOPs
m: [0.67, 0.75, 768] # YOLOv8m-ghost summary: 603 layers, 10336312 parameters, 10336296 gradients, 32.7 GFLOPs
l: [1.00, 1.00, 512] # YOLOv8l-ghost summary: 803 layers, 14277872 parameters, 14277856 gradients, 53.7 GFLOPs
x: [1.00, 1.25, 512] # YOLOv8x-ghost summary: 803 layers, 22229308 parameters, 22229292 gradients, 83.3 GFLOPs
# YOLOv8.0n-ghost backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, GhostConv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C3Ghost, [128, True]]
- [-1, 1, GhostConv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C3Ghost, [256, True]]
- [-1, 1, GhostConv, [512, 3, 2]] # 5-P4/16
- [-1, 6, C3Ghost, [512, True]]
- [-1, 1, GhostConv, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C3Ghost, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C3Ghost, [512]] # 12
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C3Ghost, [256]] # 15 (P3/8-small)
- [-1, 1, GhostConv, [256, 3, 2]]
- [[-1, 12], 1, Concat, [1]] # cat head P4
- [-1, 3, C3Ghost, [512]] # 18 (P4/16-medium)
- [-1, 1, GhostConv, [512, 3, 2]]
- [[-1, 9], 1, Concat, [1]] # cat head P5
- [-1, 3, C3Ghost, [1024]] # 21 (P5/32-large)
- [[15, 18, 21], 1, Detect, [nc]] # Detect(P3, P4, P5)

46
ultralytics/cfg/models/v8/yolov8-obb.yaml
View File

@ -1,46 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8 Oriented Bounding Boxes (OBB) model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024] # YOLOv8n summary: 225 layers, 3157200 parameters, 3157184 gradients, 8.9 GFLOPs
s: [0.33, 0.50, 1024] # YOLOv8s summary: 225 layers, 11166560 parameters, 11166544 gradients, 28.8 GFLOPs
m: [0.67, 0.75, 768] # YOLOv8m summary: 295 layers, 25902640 parameters, 25902624 gradients, 79.3 GFLOPs
l: [1.00, 1.00, 512] # YOLOv8l summary: 365 layers, 43691520 parameters, 43691504 gradients, 165.7 GFLOPs
x: [1.00, 1.25, 512] # YOLOv8x summary: 365 layers, 68229648 parameters, 68229632 gradients, 258.5 GFLOPs
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2f, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2f, [512]] # 12
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2f, [256]] # 15 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 12], 1, Concat, [1]] # cat head P4
- [-1, 3, C2f, [512]] # 18 (P4/16-medium)
- [-1, 1, Conv, [512, 3, 2]]
- [[-1, 9], 1, Concat, [1]] # cat head P5
- [-1, 3, C2f, [1024]] # 21 (P5/32-large)
- [[15, 18, 21], 1, OBB, [nc, 1]] # OBB(P3, P4, P5)

54
ultralytics/cfg/models/v8/yolov8-p2.yaml
View File

@ -1,54 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8 object detection model with P2-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024]
s: [0.33, 0.50, 1024]
m: [0.67, 0.75, 768]
l: [1.00, 1.00, 512]
x: [1.00, 1.25, 512]
# YOLOv8.0 backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2f, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
# YOLOv8.0-p2 head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2f, [512]] # 12
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2f, [256]] # 15 (P3/8-small)
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 2], 1, Concat, [1]] # cat backbone P2
- [-1, 3, C2f, [128]] # 18 (P2/4-xsmall)
- [-1, 1, Conv, [128, 3, 2]]
- [[-1, 15], 1, Concat, [1]] # cat head P3
- [-1, 3, C2f, [256]] # 21 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 12], 1, Concat, [1]] # cat head P4
- [-1, 3, C2f, [512]] # 24 (P4/16-medium)
- [-1, 1, Conv, [512, 3, 2]]
- [[-1, 9], 1, Concat, [1]] # cat head P5
- [-1, 3, C2f, [1024]] # 27 (P5/32-large)
- [[18, 21, 24, 27], 1, Detect, [nc]] # Detect(P2, P3, P4, P5)

56
ultralytics/cfg/models/v8/yolov8-p6.yaml
View File

@ -1,56 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8 object detection model with P3-P6 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n-p6.yaml' will call yolov8-p6.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024]
s: [0.33, 0.50, 1024]
m: [0.67, 0.75, 768]
l: [1.00, 1.00, 512]
x: [1.00, 1.25, 512]
# YOLOv8.0x6 backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, Conv, [768, 3, 2]] # 7-P5/32
- [-1, 3, C2f, [768, True]]
- [-1, 1, Conv, [1024, 3, 2]] # 9-P6/64
- [-1, 3, C2f, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 11
# YOLOv8.0x6 head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 8], 1, Concat, [1]] # cat backbone P5
- [-1, 3, C2, [768, False]] # 14
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2, [512, False]] # 17
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2, [256, False]] # 20 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 17], 1, Concat, [1]] # cat head P4
- [-1, 3, C2, [512, False]] # 23 (P4/16-medium)
- [-1, 1, Conv, [512, 3, 2]]
- [[-1, 14], 1, Concat, [1]] # cat head P5
- [-1, 3, C2, [768, False]] # 26 (P5/32-large)
- [-1, 1, Conv, [768, 3, 2]]
- [[-1, 11], 1, Concat, [1]] # cat head P6
- [-1, 3, C2, [1024, False]] # 29 (P6/64-xlarge)
- [[20, 23, 26, 29], 1, Detect, [nc]] # Detect(P3, P4, P5, P6)

57
ultralytics/cfg/models/v8/yolov8-pose-p6.yaml
View File

@ -1,57 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8-pose-p6 keypoints/pose estimation model. For Usage examples see https://docs.ultralytics.com/tasks/pose
# Parameters
nc: 1 # number of classes
kpt_shape: [17, 3] # number of keypoints, number of dims (2 for x,y or 3 for x,y,visible)
scales: # model compound scaling constants, i.e. 'model=yolov8n-p6.yaml' will call yolov8-p6.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024]
s: [0.33, 0.50, 1024]
m: [0.67, 0.75, 768]
l: [1.00, 1.00, 512]
x: [1.00, 1.25, 512]
# YOLOv8.0x6 backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, Conv, [768, 3, 2]] # 7-P5/32
- [-1, 3, C2f, [768, True]]
- [-1, 1, Conv, [1024, 3, 2]] # 9-P6/64
- [-1, 3, C2f, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 11
# YOLOv8.0x6 head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 8], 1, Concat, [1]] # cat backbone P5
- [-1, 3, C2, [768, False]] # 14
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2, [512, False]] # 17
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2, [256, False]] # 20 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 17], 1, Concat, [1]] # cat head P4
- [-1, 3, C2, [512, False]] # 23 (P4/16-medium)
- [-1, 1, Conv, [512, 3, 2]]
- [[-1, 14], 1, Concat, [1]] # cat head P5
- [-1, 3, C2, [768, False]] # 26 (P5/32-large)
- [-1, 1, Conv, [768, 3, 2]]
- [[-1, 11], 1, Concat, [1]] # cat head P6
- [-1, 3, C2, [1024, False]] # 29 (P6/64-xlarge)
- [[20, 23, 26, 29], 1, Pose, [nc, kpt_shape]] # Pose(P3, P4, P5, P6)

47
ultralytics/cfg/models/v8/yolov8-pose.yaml
View File

@ -1,47 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8-pose keypoints/pose estimation model. For Usage examples see https://docs.ultralytics.com/tasks/pose
# Parameters
nc: 1 # number of classes
kpt_shape: [17, 3] # number of keypoints, number of dims (2 for x,y or 3 for x,y,visible)
scales: # model compound scaling constants, i.e. 'model=yolov8n-pose.yaml' will call yolov8-pose.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024]
s: [0.33, 0.50, 1024]
m: [0.67, 0.75, 768]
l: [1.00, 1.00, 512]
x: [1.00, 1.25, 512]
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2f, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2f, [512]] # 12
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2f, [256]] # 15 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 12], 1, Concat, [1]] # cat head P4
- [-1, 3, C2f, [512]] # 18 (P4/16-medium)
- [-1, 1, Conv, [512, 3, 2]]
- [[-1, 9], 1, Concat, [1]] # cat head P5
- [-1, 3, C2f, [1024]] # 21 (P5/32-large)
- [[15, 18, 21], 1, Pose, [nc, kpt_shape]] # Pose(P3, P4, P5)

46
ultralytics/cfg/models/v8/yolov8-rtdetr.yaml
View File

@ -1,46 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8 object detection model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024] # YOLOv8n summary: 225 layers, 3157200 parameters, 3157184 gradients, 8.9 GFLOPs
s: [0.33, 0.50, 1024] # YOLOv8s summary: 225 layers, 11166560 parameters, 11166544 gradients, 28.8 GFLOPs
m: [0.67, 0.75, 768] # YOLOv8m summary: 295 layers, 25902640 parameters, 25902624 gradients, 79.3 GFLOPs
l: [1.00, 1.00, 512] # YOLOv8l summary: 365 layers, 43691520 parameters, 43691504 gradients, 165.7 GFLOPs
x: [1.00, 1.25, 512] # YOLOv8x summary: 365 layers, 68229648 parameters, 68229632 gradients, 258.5 GFLOPs
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2f, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2f, [512]] # 12
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2f, [256]] # 15 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 12], 1, Concat, [1]] # cat head P4
- [-1, 3, C2f, [512]] # 18 (P4/16-medium)
- [-1, 1, Conv, [512, 3, 2]]
- [[-1, 9], 1, Concat, [1]] # cat head P5
- [-1, 3, C2f, [1024]] # 21 (P5/32-large)
- [[15, 18, 21], 1, RTDETRDecoder, [nc]] # Detect(P3, P4, P5)

56
ultralytics/cfg/models/v8/yolov8-seg-p6.yaml
View File

@ -1,56 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8-seg-p6 instance segmentation model. For Usage examples see https://docs.ultralytics.com/tasks/segment
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n-seg-p6.yaml' will call yolov8-seg-p6.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024]
s: [0.33, 0.50, 1024]
m: [0.67, 0.75, 768]
l: [1.00, 1.00, 512]
x: [1.00, 1.25, 512]
# YOLOv8.0x6 backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, Conv, [768, 3, 2]] # 7-P5/32
- [-1, 3, C2f, [768, True]]
- [-1, 1, Conv, [1024, 3, 2]] # 9-P6/64
- [-1, 3, C2f, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 11
# YOLOv8.0x6 head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 8], 1, Concat, [1]] # cat backbone P5
- [-1, 3, C2, [768, False]] # 14
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2, [512, False]] # 17
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2, [256, False]] # 20 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 17], 1, Concat, [1]] # cat head P4
- [-1, 3, C2, [512, False]] # 23 (P4/16-medium)
- [-1, 1, Conv, [512, 3, 2]]
- [[-1, 14], 1, Concat, [1]] # cat head P5
- [-1, 3, C2, [768, False]] # 26 (P5/32-large)
- [-1, 1, Conv, [768, 3, 2]]
- [[-1, 11], 1, Concat, [1]] # cat head P6
- [-1, 3, C2, [1024, False]] # 29 (P6/64-xlarge)
- [[20, 23, 26, 29], 1, Segment, [nc, 32, 256]] # Pose(P3, P4, P5, P6)

46
ultralytics/cfg/models/v8/yolov8-seg.yaml
View File

@ -1,46 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8-seg instance segmentation model. For Usage examples see https://docs.ultralytics.com/tasks/segment
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n-seg.yaml' will call yolov8-seg.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024]
s: [0.33, 0.50, 1024]
m: [0.67, 0.75, 768]
l: [1.00, 1.00, 512]
x: [1.00, 1.25, 512]
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2f, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2f, [512]] # 12
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2f, [256]] # 15 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 12], 1, Concat, [1]] # cat head P4
- [-1, 3, C2f, [512]] # 18 (P4/16-medium)
- [-1, 1, Conv, [512, 3, 2]]
- [[-1, 9], 1, Concat, [1]] # cat head P5
- [-1, 3, C2f, [1024]] # 21 (P5/32-large)
- [[15, 18, 21], 1, Segment, [nc, 32, 256]] # Segment(P3, P4, P5)

48
ultralytics/cfg/models/v8/yolov8-world.yaml
View File

@ -1,48 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8-World object detection model with P3-P5 outputs. For details see https://docs.ultralytics.com/tasks/detect
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024] # YOLOv8n summary: 225 layers, 3157200 parameters, 3157184 gradients, 8.9 GFLOPs
s: [0.33, 0.50, 1024] # YOLOv8s summary: 225 layers, 11166560 parameters, 11166544 gradients, 28.8 GFLOPs
m: [0.67, 0.75, 768] # YOLOv8m summary: 295 layers, 25902640 parameters, 25902624 gradients, 79.3 GFLOPs
l: [1.00, 1.00, 512] # YOLOv8l summary: 365 layers, 43691520 parameters, 43691504 gradients, 165.7 GFLOPs
x: [1.00, 1.25, 512] # YOLOv8x summary: 365 layers, 68229648 parameters, 68229632 gradients, 258.5 GFLOPs
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2f, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2fAttn, [512, 256, 8]] # 12
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2fAttn, [256, 128, 4]] # 15 (P3/8-small)
- [[15, 12, 9], 1, ImagePoolingAttn, [256]] # 16 (P3/8-small)
- [15, 1, Conv, [256, 3, 2]]
- [[-1, 12], 1, Concat, [1]] # cat head P4
- [-1, 3, C2fAttn, [512, 256, 8]] # 19 (P4/16-medium)
- [-1, 1, Conv, [512, 3, 2]]
- [[-1, 9], 1, Concat, [1]] # cat head P5
- [-1, 3, C2fAttn, [1024, 512, 16]] # 22 (P5/32-large)
- [[15, 19, 22], 1, WorldDetect, [nc, 512, False]] # Detect(P3, P4, P5)

46
ultralytics/cfg/models/v8/yolov8-worldv2.yaml
View File

@ -1,46 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8-World-v2 object detection model with P3-P5 outputs. For details see https://docs.ultralytics.com/tasks/detect
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024] # YOLOv8n summary: 225 layers, 3157200 parameters, 3157184 gradients, 8.9 GFLOPs
s: [0.33, 0.50, 1024] # YOLOv8s summary: 225 layers, 11166560 parameters, 11166544 gradients, 28.8 GFLOPs
m: [0.67, 0.75, 768] # YOLOv8m summary: 295 layers, 25902640 parameters, 25902624 gradients, 79.3 GFLOPs
l: [1.00, 1.00, 512] # YOLOv8l summary: 365 layers, 43691520 parameters, 43691504 gradients, 165.7 GFLOPs
x: [1.00, 1.25, 512] # YOLOv8x summary: 365 layers, 68229648 parameters, 68229632 gradients, 258.5 GFLOPs
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2f, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2fAttn, [512, 256, 8]] # 12
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2fAttn, [256, 128, 4]] # 15 (P3/8-small)
- [15, 1, Conv, [256, 3, 2]]
- [[-1, 12], 1, Concat, [1]] # cat head P4
- [-1, 3, C2fAttn, [512, 256, 8]] # 18 (P4/16-medium)
- [-1, 1, Conv, [512, 3, 2]]
- [[-1, 9], 1, Concat, [1]] # cat head P5
- [-1, 3, C2fAttn, [1024, 512, 16]] # 21 (P5/32-large)
- [[15, 18, 21], 1, WorldDetect, [nc, 512, True]] # Detect(P3, P4, P5)

46
ultralytics/cfg/models/v8/yolov8.yaml
View File

@ -1,46 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8 object detection model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
# Parameters
nc: 80 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
# [depth, width, max_channels]
n: [0.33, 0.25, 1024] # YOLOv8n summary: 225 layers, 3157200 parameters, 3157184 gradients, 8.9 GFLOPs
s: [0.33, 0.50, 1024] # YOLOv8s summary: 225 layers, 11166560 parameters, 11166544 gradients, 28.8 GFLOPs
m: [0.67, 0.75, 768] # YOLOv8m summary: 295 layers, 25902640 parameters, 25902624 gradients, 79.3 GFLOPs
l: [1.00, 1.00, 512] # YOLOv8l summary: 365 layers, 43691520 parameters, 43691504 gradients, 165.7 GFLOPs
x: [1.00, 1.25, 512] # YOLOv8x summary: 365 layers, 68229648 parameters, 68229632 gradients, 258.5 GFLOPs
# YOLOv8.0n backbone
backbone:
# [from, repeats, module, args]
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 3, C2f, [128, True]]
- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
- [-1, 6, C2f, [256, True]]
- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
- [-1, 6, C2f, [512, True]]
- [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32
- [-1, 3, C2f, [1024, True]]
- [-1, 1, SPPF, [1024, 5]] # 9
# YOLOv8.0n head
head:
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 3, C2f, [512]] # 12
- [-1, 1, nn.Upsample, [None, 2, "nearest"]]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 3, C2f, [256]] # 15 (P3/8-small)
- [-1, 1, Conv, [256, 3, 2]]
- [[-1, 12], 1, Concat, [1]] # cat head P4
- [-1, 3, C2f, [512]] # 18 (P4/16-medium)
- [-1, 1, Conv, [512, 3, 2]]
- [[-1, 9], 1, Concat, [1]] # cat head P5
- [-1, 3, C2f, [1024]] # 21 (P5/32-large)
- [[15, 18, 21], 1, Detect, [nc]] # Detect(P3, P4, P5)

36
ultralytics/cfg/models/v9/yolov9c.yaml
View File

@ -1,36 +0,0 @@
# YOLOv9
# parameters
nc: 80 # number of classes
# gelan backbone
backbone:
- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
- [-1, 1, RepNCSPELAN4, [256, 128, 64, 1]] # 2
- [-1, 1, ADown, [256]] # 3-P3/8
- [-1, 1, RepNCSPELAN4, [512, 256, 128, 1]] # 4
- [-1, 1, ADown, [512]] # 5-P4/16
- [-1, 1, RepNCSPELAN4, [512, 512, 256, 1]] # 6
- [-1, 1, ADown, [512]] # 7-P5/32
- [-1, 1, RepNCSPELAN4, [512, 512, 256, 1]] # 8
- [-1, 1, SPPELAN, [512, 256]] # 9
head:
- [-1, 1, nn.Upsample, [None, 2, 'nearest']]
- [[-1, 6], 1, Concat, [1]] # cat backbone P4
- [-1, 1, RepNCSPELAN4, [512, 512, 256, 1]] # 12
- [-1, 1, nn.Upsample, [None, 2, 'nearest']]
- [[-1, 4], 1, Concat, [1]] # cat backbone P3
- [-1, 1, RepNCSPELAN4, [256, 256, 128, 1]] # 15 (P3/8-small)
- [-1, 1, ADown, [256]]
- [[-1, 12], 1, Concat, [1]] # cat head P4
- [-1, 1, RepNCSPELAN4, [512, 512, 256, 1]] # 18 (P4/16-medium)
- [-1, 1, ADown, [512]]
- [[-1, 9], 1, Concat, [1]] # cat head P5
- [-1, 1, RepNCSPELAN4, [512, 512, 256, 1]] # 21 (P5/32-large)
- [[15, 18, 21], 1, Detect, [nc]] # DDetect(P3, P4, P5)

60
ultralytics/cfg/models/v9/yolov9e.yaml
View File

@ -1,60 +0,0 @@
# YOLOv9
# parameters
nc: 80 # number of classes
# gelan backbone
backbone:
- [-1, 1, Silence, []]
- [-1, 1, Conv, [64, 3, 2]] # 1-P1/2
- [-1, 1, Conv, [128, 3, 2]] # 2-P2/4
- [-1, 1, RepNCSPELAN4, [256, 128, 64, 2]] # 3
- [-1, 1, ADown, [256]] # 4-P3/8
- [-1, 1, RepNCSPELAN4, [512, 256, 128, 2]] # 5
- [-1, 1, ADown, [512]] # 6-P4/16
- [-1, 1, RepNCSPELAN4, [1024, 512, 256, 2]] # 7
- [-1, 1, ADown, [1024]] # 8-P5/32
- [-1, 1, RepNCSPELAN4, [1024, 512, 256, 2]] # 9
- [1, 1, CBLinear, [[64]]] # 10
- [3, 1, CBLinear, [[64, 128]]] # 11
- [5, 1, CBLinear, [[64, 128, 256]]] # 12
- [7, 1, CBLinear, [[64, 128, 256, 512]]] # 13
- [9, 1, CBLinear, [[64, 128, 256, 512, 1024]]] # 14
- [0, 1, Conv, [64, 3, 2]] # 15-P1/2
- [[10, 11, 12, 13, 14, -1], 1, CBFuse, [[0, 0, 0, 0, 0]]] # 16
- [-1, 1, Conv, [128, 3, 2]] # 17-P2/4
- [[11, 12, 13, 14, -1], 1, CBFuse, [[1, 1, 1, 1]]] # 18
- [-1, 1, RepNCSPELAN4, [256, 128, 64, 2]] # 19
- [-1, 1, ADown, [256]] # 20-P3/8
- [[12, 13, 14, -1], 1, CBFuse, [[2, 2, 2]]] # 21
- [-1, 1, RepNCSPELAN4, [512, 256, 128, 2]] # 22
- [-1, 1, ADown, [512]] # 23-P4/16
- [[13, 14, -1], 1, CBFuse, [[3, 3]]] # 24
- [-1, 1, RepNCSPELAN4, [1024, 512, 256, 2]] # 25
- [-1, 1, ADown, [1024]] # 26-P5/32
- [[14, -1], 1, CBFuse, [[4]]] # 27
- [-1, 1, RepNCSPELAN4, [1024, 512, 256, 2]] # 28
- [-1, 1, SPPELAN, [512, 256]] # 29
# gelan head
head:
- [-1, 1, nn.Upsample, [None, 2, 'nearest']]
- [[-1, 25], 1, Concat, [1]] # cat backbone P4
- [-1, 1, RepNCSPELAN4, [512, 512, 256, 2]] # 32
- [-1, 1, nn.Upsample, [None, 2, 'nearest']]
- [[-1, 22], 1, Concat, [1]] # cat backbone P3
- [-1, 1, RepNCSPELAN4, [256, 256, 128, 2]] # 35 (P3/8-small)
- [-1, 1, ADown, [256]]
- [[-1, 32], 1, Concat, [1]] # cat head P4
- [-1, 1, RepNCSPELAN4, [512, 512, 256, 2]] # 38 (P4/16-medium)
- [-1, 1, ADown, [512]]
- [[-1, 29], 1, Concat, [1]] # cat head P5
- [-1, 1, RepNCSPELAN4, [512, 1024, 512, 2]] # 41 (P5/32-large)
# detect
- [[35, 38, 41], 1, Detect, [nc]] # Detect(P3, P4, P5)

18
ultralytics/cfg/trackers/botsort.yaml
View File

@ -1,18 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# Default YOLO tracker settings for BoT-SORT tracker https://github.com/NirAharon/BoT-SORT
tracker_type: botsort # tracker type, ['botsort', 'bytetrack']
track_high_thresh: 0.5 # threshold for the first association
track_low_thresh: 0.1 # threshold for the second association
new_track_thresh: 0.6 # threshold for init new track if the detection does not match any tracks
track_buffer: 30 # buffer to calculate the time when to remove tracks
match_thresh: 0.8 # threshold for matching tracks
# min_box_area: 10 # threshold for min box areas(for tracker evaluation, not used for now)
# mot20: False # for tracker evaluation(not used for now)
# BoT-SORT settings
gmc_method: sparseOptFlow # method of global motion compensation
# ReID model related thresh (not supported yet)
proximity_thresh: 0.5
appearance_thresh: 0.25
with_reid: False

11
ultralytics/cfg/trackers/bytetrack.yaml
View File

@ -1,11 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# Default YOLO tracker settings for ByteTrack tracker https://github.com/ifzhang/ByteTrack
tracker_type: bytetrack # tracker type, ['botsort', 'bytetrack']
track_high_thresh: 0.5 # threshold for the first association
track_low_thresh: 0.1 # threshold for the second association
new_track_thresh: 0.6 # threshold for init new track if the detection does not match any tracks
track_buffer: 30 # buffer to calculate the time when to remove tracks
match_thresh: 0.8 # threshold for matching tracks
# min_box_area: 10 # threshold for min box areas(for tracker evaluation, not used for now)
# mot20: False # for tracker evaluation(not used for now)

15
ultralytics/data/__init__.py
View File

@ -1,15 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
from .base import BaseDataset
from .build import build_dataloader, build_yolo_dataset, load_inference_source
from .dataset import ClassificationDataset, SemanticDataset, YOLODataset
__all__ = (
"BaseDataset",
"ClassificationDataset",
"SemanticDataset",
"YOLODataset",
"build_yolo_dataset",
"build_dataloader",
"load_inference_source",
)

50
ultralytics/data/annotator.py
View File

@ -1,50 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
from pathlib import Path
from ultralytics import SAM, YOLO
def auto_annotate(data, det_model="yolov8x.pt", sam_model="sam_b.pt", device="", output_dir=None):
"""
Automatically annotates images using a YOLO object detection model and a SAM segmentation model.
Args:
data (str): Path to a folder containing images to be annotated.
det_model (str, optional): Pre-trained YOLO detection model. Defaults to 'yolov8x.pt'.
sam_model (str, optional): Pre-trained SAM segmentation model. Defaults to 'sam_b.pt'.
device (str, optional): Device to run the models on. Defaults to an empty string (CPU or GPU, if available).
output_dir (str | None | optional): Directory to save the annotated results.
Defaults to a 'labels' folder in the same directory as 'data'.
Example:
```python
from ultralytics.data.annotator import auto_annotate
auto_annotate(data='ultralytics/assets', det_model='yolov8n.pt', sam_model='mobile_sam.pt')
```
"""
det_model = YOLO(det_model)
sam_model = SAM(sam_model)
data = Path(data)
if not output_dir:
output_dir = data.parent / f"{data.stem}_auto_annotate_labels"
Path(output_dir).mkdir(exist_ok=True, parents=True)
det_results = det_model(data, stream=True, device=device)
for result in det_results:
class_ids = result.boxes.cls.int().tolist() # noqa
if len(class_ids):
boxes = result.boxes.xyxy # Boxes object for bbox outputs
sam_results = sam_model(result.orig_img, bboxes=boxes, verbose=False, save=False, device=device)
segments = sam_results[0].masks.xyn # noqa
with open(f"{Path(output_dir) / Path(result.path).stem}.txt", "w") as f:
for i in range(len(segments)):
s = segments[i]
if len(s) == 0:
continue
segment = map(str, segments[i].reshape(-1).tolist())
f.write(f"{class_ids[i]} " + " ".join(segment) + "\n")

1254
ultralytics/data/augment.py
View File

File diff suppressed because it is too large Load Diff

311
ultralytics/data/base.py
View File

@ -1,311 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import glob
import math
import os
import random
from copy import deepcopy
from multiprocessing.pool import ThreadPool
from pathlib import Path
from typing import Optional
import cv2
import numpy as np
import psutil
from torch.utils.data import Dataset
from ultralytics.utils import DEFAULT_CFG, LOCAL_RANK, LOGGER, NUM_THREADS, TQDM
from .utils import HELP_URL, IMG_FORMATS
class BaseDataset(Dataset):
"""
Base dataset class for loading and processing image data.
Args:
img_path (str): Path to the folder containing images.
imgsz (int, optional): Image size. Defaults to 640.
cache (bool, optional): Cache images to RAM or disk during training. Defaults to False.
augment (bool, optional): If True, data augmentation is applied. Defaults to True.
hyp (dict, optional): Hyperparameters to apply data augmentation. Defaults to None.
prefix (str, optional): Prefix to print in log messages. Defaults to ''.
rect (bool, optional): If True, rectangular training is used. Defaults to False.
batch_size (int, optional): Size of batches. Defaults to None.
stride (int, optional): Stride. Defaults to 32.
pad (float, optional): Padding. Defaults to 0.0.
single_cls (bool, optional): If True, single class training is used. Defaults to False.
classes (list): List of included classes. Default is None.
fraction (float): Fraction of dataset to utilize. Default is 1.0 (use all data).
Attributes:
im_files (list): List of image file paths.
labels (list): List of label data dictionaries.
ni (int): Number of images in the dataset.
ims (list): List of loaded images.
npy_files (list): List of numpy file paths.
transforms (callable): Image transformation function.
"""
def __init__(
self,
img_path,
imgsz=640,
cache=False,
augment=True,
hyp=DEFAULT_CFG,
prefix="",
rect=False,
batch_size=16,
stride=32,
pad=0.5,
single_cls=False,
classes=None,
fraction=1.0,
):
"""Initialize BaseDataset with given configuration and options."""
super().__init__()
self.img_path = img_path
self.imgsz = imgsz
self.augment = augment
self.single_cls = single_cls
self.prefix = prefix
self.fraction = fraction
self.im_files = self.get_img_files(self.img_path)
self.labels = self.get_labels()
self.update_labels(include_class=classes) # single_cls and include_class
self.ni = len(self.labels) # number of images
self.rect = rect
self.batch_size = batch_size
self.stride = stride
self.pad = pad
if self.rect:
assert self.batch_size is not None
self.set_rectangle()
# Buffer thread for mosaic images
self.buffer = [] # buffer size = batch size
self.max_buffer_length = min((self.ni, self.batch_size * 8, 1000)) if self.augment else 0
# Cache images
if cache == "ram" and not self.check_cache_ram():
cache = False
self.ims, self.im_hw0, self.im_hw = [None] * self.ni, [None] * self.ni, [None] * self.ni
self.npy_files = [Path(f).with_suffix(".npy") for f in self.im_files]
if cache:
self.cache_images(cache)
# Transforms
self.transforms = self.build_transforms(hyp=hyp)
def get_img_files(self, img_path):
"""Read image files."""
try:
f = [] # image files
for p in img_path if isinstance(img_path, list) else [img_path]:
p = Path(p) # os-agnostic
if p.is_dir(): # dir
f += glob.glob(str(p / "**" / "*.*"), recursive=True)
# F = list(p.rglob('*.*')) # pathlib
elif p.is_file(): # file
with open(p) as t:
t = t.read().strip().splitlines()
parent = str(p.parent) + os.sep
f += [x.replace("./", parent) if x.startswith("./") else x for x in t] # local to global path
# F += [p.parent / x.lstrip(os.sep) for x in t] # local to global path (pathlib)
else:
raise FileNotFoundError(f"{self.prefix}{p} does not exist")
im_files = sorted(x.replace("/", os.sep) for x in f if x.split(".")[-1].lower() in IMG_FORMATS)
# self.img_files = sorted([x for x in f if x.suffix[1:].lower() in IMG_FORMATS]) # pathlib
assert im_files, f"{self.prefix}No images found in {img_path}"
except Exception as e:
raise FileNotFoundError(f"{self.prefix}Error loading data from {img_path}\n{HELP_URL}") from e
if self.fraction < 1:
# im_files = im_files[: round(len(im_files) * self.fraction)]
num_elements_to_select = round(len(im_files) * self.fraction)
im_files = random.sample(im_files, num_elements_to_select)
return im_files
def update_labels(self, include_class: Optional[list]):
"""Update labels to include only these classes (optional)."""
include_class_array = np.array(include_class).reshape(1, -1)
for i in range(len(self.labels)):
if include_class is not None:
cls = self.labels[i]["cls"]
bboxes = self.labels[i]["bboxes"]
segments = self.labels[i]["segments"]
keypoints = self.labels[i]["keypoints"]
j = (cls == include_class_array).any(1)
self.labels[i]["cls"] = cls[j]
self.labels[i]["bboxes"] = bboxes[j]
if segments:
self.labels[i]["segments"] = [segments[si] for si, idx in enumerate(j) if idx]
if keypoints is not None:
self.labels[i]["keypoints"] = keypoints[j]
if self.single_cls:
self.labels[i]["cls"][:, 0] = 0
def load_image(self, i, rect_mode=True):
"""Loads 1 image from dataset index 'i', returns (im, resized hw)."""
im, f, fn = self.ims[i], self.im_files[i], self.npy_files[i]
if im is None: # not cached in RAM
if fn.exists(): # load npy
try:
im = np.load(fn)
except Exception as e:
LOGGER.warning(f"{self.prefix}WARNING ⚠️ Removing corrupt *.npy image file {fn} due to: {e}")
Path(fn).unlink(missing_ok=True)
im = cv2.imread(f) # BGR
else: # read image
im = cv2.imread(f) # BGR
if im is None:
raise FileNotFoundError(f"Image Not Found {f}")
h0, w0 = im.shape[:2] # orig hw
if rect_mode: # resize long side to imgsz while maintaining aspect ratio
r = self.imgsz / max(h0, w0) # ratio
if r != 1: # if sizes are not equal
w, h = (min(math.ceil(w0 * r), self.imgsz), min(math.ceil(h0 * r), self.imgsz))
im = cv2.resize(im, (w, h), interpolation=cv2.INTER_LINEAR)
elif not (h0 == w0 == self.imgsz): # resize by stretching image to square imgsz
im = cv2.resize(im, (self.imgsz, self.imgsz), interpolation=cv2.INTER_LINEAR)
# Add to buffer if training with augmentations
if self.augment:
self.ims[i], self.im_hw0[i], self.im_hw[i] = im, (h0, w0), im.shape[:2] # im, hw_original, hw_resized
self.buffer.append(i)
if len(self.buffer) >= self.max_buffer_length:
j = self.buffer.pop(0)
self.ims[j], self.im_hw0[j], self.im_hw[j] = None, None, None
return im, (h0, w0), im.shape[:2]
return self.ims[i], self.im_hw0[i], self.im_hw[i]
def cache_images(self, cache):
"""Cache images to memory or disk."""
b, gb = 0, 1 << 30 # bytes of cached images, bytes per gigabytes
fcn = self.cache_images_to_disk if cache == "disk" else self.load_image
with ThreadPool(NUM_THREADS) as pool:
results = pool.imap(fcn, range(self.ni))
pbar = TQDM(enumerate(results), total=self.ni, disable=LOCAL_RANK > 0)
for i, x in pbar:
if cache == "disk":
b += self.npy_files[i].stat().st_size
else: # 'ram'
self.ims[i], self.im_hw0[i], self.im_hw[i] = x # im, hw_orig, hw_resized = load_image(self, i)
b += self.ims[i].nbytes
pbar.desc = f"{self.prefix}Caching images ({b / gb:.1f}GB {cache})"
pbar.close()
def cache_images_to_disk(self, i):
"""Saves an image as an *.npy file for faster loading."""
f = self.npy_files[i]
if not f.exists():
np.save(f.as_posix(), cv2.imread(self.im_files[i]), allow_pickle=False)
def check_cache_ram(self, safety_margin=0.5):
"""Check image caching requirements vs available memory."""
b, gb = 0, 1 << 30 # bytes of cached images, bytes per gigabytes
n = min(self.ni, 30) # extrapolate from 30 random images
for _ in range(n):
im = cv2.imread(random.choice(self.im_files)) # sample image
ratio = self.imgsz / max(im.shape[0], im.shape[1]) # max(h, w) # ratio
b += im.nbytes * ratio**2
mem_required = b * self.ni / n * (1 + safety_margin) # GB required to cache dataset into RAM
mem = psutil.virtual_memory()
cache = mem_required < mem.available # to cache or not to cache, that is the question
if not cache:
LOGGER.info(
f'{self.prefix}{mem_required / gb:.1f}GB RAM required to cache images '
f'with {int(safety_margin * 100)}% safety margin but only '
f'{mem.available / gb:.1f}/{mem.total / gb:.1f}GB available, '
f"{'caching images ✅' if cache else 'not caching images ⚠️'}"
)
return cache
def set_rectangle(self):
"""Sets the shape of bounding boxes for YOLO detections as rectangles."""
bi = np.floor(np.arange(self.ni) / self.batch_size).astype(int) # batch index
nb = bi[-1] + 1 # number of batches
s = np.array([x.pop("shape") for x in self.labels]) # hw
ar = s[:, 0] / s[:, 1] # aspect ratio
irect = ar.argsort()
self.im_files = [self.im_files[i] for i in irect]
self.labels = [self.labels[i] for i in irect]
ar = ar[irect]
# Set training image shapes
shapes = [[1, 1]] * nb
for i in range(nb):
ari = ar[bi == i]
mini, maxi = ari.min(), ari.max()
if maxi < 1:
shapes[i] = [maxi, 1]
elif mini > 1:
shapes[i] = [1, 1 / mini]
self.batch_shapes = np.ceil(np.array(shapes) * self.imgsz / self.stride + self.pad).astype(int) * self.stride
self.batch = bi # batch index of image
def __getitem__(self, index):
"""Returns transformed label information for given index."""
return self.transforms(self.get_image_and_label(index))
def get_image_and_label(self, index):
"""Get and return label information from the dataset."""
label = deepcopy(self.labels[index]) # requires deepcopy() https://github.com/ultralytics/ultralytics/pull/1948
label.pop("shape", None) # shape is for rect, remove it
label["img"], label["ori_shape"], label["resized_shape"] = self.load_image(index)
label["ratio_pad"] = (
label["resized_shape"][0] / label["ori_shape"][0],
label["resized_shape"][1] / label["ori_shape"][1],
) # for evaluation
if self.rect:
label["rect_shape"] = self.batch_shapes[self.batch[index]]
return self.update_labels_info(label)
def __len__(self):
"""Returns the length of the labels list for the dataset."""
return int(0.1*len(self.labels))
def update_labels_info(self, label):
"""Custom your label format here."""
return label
def build_transforms(self, hyp=None):
"""
Users can customize augmentations here.
Example:
```python
if self.augment:
# Training transforms
return Compose([])
else:
# Val transforms
return Compose([])
```
"""
raise NotImplementedError
def get_labels(self):
"""
Users can customize their own format here.
Note:
Ensure output is a dictionary with the following keys:
```python
dict(
im_file=im_file,
shape=shape, # format: (height, width)
cls=cls,
bboxes=bboxes, # xywh
segments=segments, # xy
keypoints=keypoints, # xy
normalized=True, # or False
bbox_format="xyxy", # or xywh, ltwh
)
```
"""
raise NotImplementedError

186
ultralytics/data/build.py
View File

@ -1,186 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import os
import random
from pathlib import Path
import numpy as np
import torch
from PIL import Image
from torch.utils.data import dataloader, distributed
from ultralytics.data.loaders import (
LOADERS,
LoadImagesAndVideos,
LoadPilAndNumpy,
LoadScreenshots,
LoadStreams,
LoadTensor,
SourceTypes,
autocast_list,
)
from ultralytics.data.utils import IMG_FORMATS, VID_FORMATS
from ultralytics.utils import RANK, colorstr
from ultralytics.utils.checks import check_file
from .dataset import YOLODataset
from .utils import PIN_MEMORY
class InfiniteDataLoader(dataloader.DataLoader):
"""
Dataloader that reuses workers.
Uses same syntax as vanilla DataLoader.
"""
def __init__(self, *args, **kwargs):
"""Dataloader that infinitely recycles workers, inherits from DataLoader."""
super().__init__(*args, **kwargs)
object.__setattr__(self, "batch_sampler", _RepeatSampler(self.batch_sampler))
self.iterator = super().__iter__()
def __len__(self):
"""Returns the length of the batch sampler's sampler."""
return len(self.batch_sampler.sampler)
def __iter__(self):
"""Creates a sampler that repeats indefinitely."""
for _ in range(len(self)):
yield next(self.iterator)
def reset(self):
"""
Reset iterator.
This is useful when we want to modify settings of dataset while training.
"""
self.iterator = self._get_iterator()
class _RepeatSampler:
"""
Sampler that repeats forever.
Args:
sampler (Dataset.sampler): The sampler to repeat.
"""
def __init__(self, sampler):
"""Initializes an object that repeats a given sampler indefinitely."""
self.sampler = sampler
def __iter__(self):
"""Iterates over the 'sampler' and yields its contents."""
while True:
yield from iter(self.sampler)
def seed_worker(worker_id): # noqa
"""Set dataloader worker seed https://pytorch.org/docs/stable/notes/randomness.html#dataloader."""
worker_seed = torch.initial_seed() % 2**32
np.random.seed(worker_seed)
random.seed(worker_seed)
def build_yolo_dataset(cfg, img_path, batch, data, mode="train", rect=False, stride=32):
"""Build YOLO Dataset."""
return YOLODataset(
img_path=img_path,
imgsz=cfg.imgsz,
batch_size=batch,
augment=mode == "train", # augmentation
hyp=cfg, # TODO: probably add a get_hyps_from_cfg function
rect=cfg.rect or rect, # rectangular batches
cache=cfg.cache or None,
single_cls=cfg.single_cls or False,
stride=int(stride),
pad=0.0 if mode == "train" else 0.5,
prefix=colorstr(f"{mode}: "),
task=cfg.task,
classes=cfg.classes,
data=data,
fraction=cfg.fraction if mode == "train" else 1.0,
)
def build_dataloader(dataset, batch, workers, shuffle=True, rank=-1):
"""Return an InfiniteDataLoader or DataLoader for training or validation set."""
batch = min(batch, len(dataset))
nd = torch.cuda.device_count() # number of CUDA devices
nw = min([os.cpu_count() // max(nd, 1), workers]) # number of workers
sampler = None if rank == -1 else distributed.DistributedSampler(dataset, shuffle=shuffle)
generator = torch.Generator()
generator.manual_seed(6148914691236517205 + RANK)
return InfiniteDataLoader(
dataset=dataset,
batch_size=batch,
shuffle=shuffle and sampler is None,
num_workers=nw,
sampler=sampler,
pin_memory=PIN_MEMORY,
collate_fn=getattr(dataset, "collate_fn", None),
worker_init_fn=seed_worker,
generator=generator,
)
def check_source(source):
"""Check source type and return corresponding flag values."""
webcam, screenshot, from_img, in_memory, tensor = False, False, False, False, False
if isinstance(source, (str, int, Path)): # int for local usb camera
source = str(source)
is_file = Path(source).suffix[1:] in (IMG_FORMATS | VID_FORMATS)
is_url = source.lower().startswith(("https://", "http://", "rtsp://", "rtmp://", "tcp://"))
webcam = source.isnumeric() or source.endswith(".streams") or (is_url and not is_file)
screenshot = source.lower() == "screen"
if is_url and is_file:
source = check_file(source) # download
elif isinstance(source, LOADERS):
in_memory = True
elif isinstance(source, (list, tuple)):
source = autocast_list(source) # convert all list elements to PIL or np arrays
from_img = True
elif isinstance(source, (Image.Image, np.ndarray)):
from_img = True
elif isinstance(source, torch.Tensor):
tensor = True
else:
raise TypeError("Unsupported image type. For supported types see https://docs.ultralytics.com/modes/predict")
return source, webcam, screenshot, from_img, in_memory, tensor
def load_inference_source(source=None, batch=1, vid_stride=1, buffer=False):
"""
Loads an inference source for object detection and applies necessary transformations.
Args:
source (str, Path, Tensor, PIL.Image, np.ndarray): The input source for inference.
batch (int, optional): Batch size for dataloaders. Default is 1.
vid_stride (int, optional): The frame interval for video sources. Default is 1.
buffer (bool, optional): Determined whether stream frames will be buffered. Default is False.
Returns:
dataset (Dataset): A dataset object for the specified input source.
"""
source, stream, screenshot, from_img, in_memory, tensor = check_source(source)
source_type = source.source_type if in_memory else SourceTypes(stream, screenshot, from_img, tensor)
# Dataloader
if tensor:
dataset = LoadTensor(source)
elif in_memory:
dataset = source
elif stream:
dataset = LoadStreams(source, vid_stride=vid_stride, buffer=buffer)
elif screenshot:
dataset = LoadScreenshots(source)
elif from_img:
dataset = LoadPilAndNumpy(source)
else:
dataset = LoadImagesAndVideos(source, batch=batch, vid_stride=vid_stride)
# Attach source types to the dataset
setattr(dataset, "source_type", source_type)
return dataset

542
ultralytics/data/converter.py
View File

@ -1,542 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import json
from collections import defaultdict
from pathlib import Path
import cv2
import numpy as np
from ultralytics.utils import LOGGER, TQDM
from ultralytics.utils.files import increment_path
def coco91_to_coco80_class():
"""
Converts 91-index COCO class IDs to 80-index COCO class IDs.
Returns:
(list): A list of 91 class IDs where the index represents the 80-index class ID and the value is the
corresponding 91-index class ID.
"""
return [
0,
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
None,
11,
12,
13,
14,
15,
16,
17,
18,
19,
20,
21,
22,
23,
None,
24,
25,
None,
None,
26,
27,
28,
29,
30,
31,
32,
33,
34,
35,
36,
37,
38,
39,
None,
40,
41,
42,
43,
44,
45,
46,
47,
48,
49,
50,
51,
52,
53,
54,
55,
56,
57,
58,
59,
None,
60,
None,
None,
61,
None,
62,
63,
64,
65,
66,
67,
68,
69,
70,
71,
72,
None,
73,
74,
75,
76,
77,
78,
79,
None,
]
def coco80_to_coco91_class():
"""
Converts 80-index (val2014) to 91-index (paper).
For details see https://tech.amikelive.com/node-718/what-object-categories-labels-are-in-coco-dataset/.
Example:
```python
import numpy as np
a = np.loadtxt('data/coco.names', dtype='str', delimiter='\n')
b = np.loadtxt('data/coco_paper.names', dtype='str', delimiter='\n')
x1 = [list(a[i] == b).index(True) + 1 for i in range(80)] # darknet to coco
x2 = [list(b[i] == a).index(True) if any(b[i] == a) else None for i in range(91)] # coco to darknet
```
"""
return [
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
13,
14,
15,
16,
17,
18,
19,
20,
21,
22,
23,
24,
25,
27,
28,
31,
32,
33,
34,
35,
36,
37,
38,
39,
40,
41,
42,
43,
44,
46,
47,
48,
49,
50,
51,
52,
53,
54,
55,
56,
57,
58,
59,
60,
61,
62,
63,
64,
65,
67,
70,
72,
73,
74,
75,
76,
77,
78,
79,
80,
81,
82,
84,
85,
86,
87,
88,
89,
90,
]
def convert_coco(
labels_dir="../coco/annotations/",
save_dir="coco_converted/",
use_segments=False,
use_keypoints=False,
cls91to80=True,
):
"""
Converts COCO dataset annotations to a YOLO annotation format suitable for training YOLO models.
Args:
labels_dir (str, optional): Path to directory containing COCO dataset annotation files.
save_dir (str, optional): Path to directory to save results to.
use_segments (bool, optional): Whether to include segmentation masks in the output.
use_keypoints (bool, optional): Whether to include keypoint annotations in the output.
cls91to80 (bool, optional): Whether to map 91 COCO class IDs to the corresponding 80 COCO class IDs.
Example:
```python
from ultralytics.data.converter import convert_coco
convert_coco('../datasets/coco/annotations/', use_segments=True, use_keypoints=False, cls91to80=True)
```
Output:
Generates output files in the specified output directory.
"""
# Create dataset directory
save_dir = increment_path(save_dir) # increment if save directory already exists
for p in save_dir / "labels", save_dir / "images":
p.mkdir(parents=True, exist_ok=True) # make dir
# Convert classes
coco80 = coco91_to_coco80_class()
# Import json
for json_file in sorted(Path(labels_dir).resolve().glob("*.json")):
fn = Path(save_dir) / "labels" / json_file.stem.replace("instances_", "") # folder name
fn.mkdir(parents=True, exist_ok=True)
with open(json_file) as f:
data = json.load(f)
# Create image dict
images = {f'{x["id"]:d}': x for x in data["images"]}
# Create image-annotations dict
imgToAnns = defaultdict(list)
for ann in data["annotations"]:
imgToAnns[ann["image_id"]].append(ann)
# Write labels file
for img_id, anns in TQDM(imgToAnns.items(), desc=f"Annotations {json_file}"):
img = images[f"{img_id:d}"]
h, w, f = img["height"], img["width"], img["file_name"]
bboxes = []
segments = []
keypoints = []
for ann in anns:
if ann["iscrowd"]:
continue
# The COCO box format is [top left x, top left y, width, height]
box = np.array(ann["bbox"], dtype=np.float64)
box[:2] += box[2:] / 2 # xy top-left corner to center
box[[0, 2]] /= w # normalize x
box[[1, 3]] /= h # normalize y
if box[2] <= 0 or box[3] <= 0: # if w <= 0 and h <= 0
continue
cls = coco80[ann["category_id"] - 1] if cls91to80 else ann["category_id"] - 1 # class
box = [cls] + box.tolist()
if box not in bboxes:
bboxes.append(box)
if use_segments and ann.get("segmentation") is not None:
if len(ann["segmentation"]) == 0:
segments.append([])
continue
elif len(ann["segmentation"]) > 1:
s = merge_multi_segment(ann["segmentation"])
s = (np.concatenate(s, axis=0) / np.array([w, h])).reshape(-1).tolist()
else:
s = [j for i in ann["segmentation"] for j in i] # all segments concatenated
s = (np.array(s).reshape(-1, 2) / np.array([w, h])).reshape(-1).tolist()
s = [cls] + s
segments.append(s)
if use_keypoints and ann.get("keypoints") is not None:
keypoints.append(
box + (np.array(ann["keypoints"]).reshape(-1, 3) / np.array([w, h, 1])).reshape(-1).tolist()
)
# Write
with open((fn / f).with_suffix(".txt"), "a") as file:
for i in range(len(bboxes)):
if use_keypoints:
line = (*(keypoints[i]),) # cls, box, keypoints
else:
line = (
*(segments[i] if use_segments and len(segments[i]) > 0 else bboxes[i]),
) # cls, box or segments
file.write(("%g " * len(line)).rstrip() % line + "\n")
LOGGER.info(f"COCO data converted successfully.\nResults saved to {save_dir.resolve()}")
def convert_dota_to_yolo_obb(dota_root_path: str):
"""
Converts DOTA dataset annotations to YOLO OBB (Oriented Bounding Box) format.
The function processes images in the 'train' and 'val' folders of the DOTA dataset. For each image, it reads the
associated label from the original labels directory and writes new labels in YOLO OBB format to a new directory.
Args:
dota_root_path (str): The root directory path of the DOTA dataset.
Example:
```python
from ultralytics.data.converter import convert_dota_to_yolo_obb
convert_dota_to_yolo_obb('path/to/DOTA')
```
Notes:
The directory structure assumed for the DOTA dataset:
- DOTA
├─ images
│ ├─ train
│ └─ val
└─ labels
├─ train_original
└─ val_original
After execution, the function will organize the labels into:
- DOTA
└─ labels
├─ train
└─ val
"""
dota_root_path = Path(dota_root_path)
# Class names to indices mapping
class_mapping = {
"plane": 0,
"ship": 1,
"storage-tank": 2,
"baseball-diamond": 3,
"tennis-court": 4,
"basketball-court": 5,
"ground-track-field": 6,
"harbor": 7,
"bridge": 8,
"large-vehicle": 9,
"small-vehicle": 10,
"helicopter": 11,
"roundabout": 12,
"soccer-ball-field": 13,
"swimming-pool": 14,
"container-crane": 15,
"airport": 16,
"helipad": 17,
}
def convert_label(image_name, image_width, image_height, orig_label_dir, save_dir):
"""Converts a single image's DOTA annotation to YOLO OBB format and saves it to a specified directory."""
orig_label_path = orig_label_dir / f"{image_name}.txt"
save_path = save_dir / f"{image_name}.txt"
with orig_label_path.open("r") as f, save_path.open("w") as g:
lines = f.readlines()
for line in lines:
parts = line.strip().split()
if len(parts) < 9:
continue
class_name = parts[8]
class_idx = class_mapping[class_name]
coords = [float(p) for p in parts[:8]]
normalized_coords = [
coords[i] / image_width if i % 2 == 0 else coords[i] / image_height for i in range(8)
]
formatted_coords = ["{:.6g}".format(coord) for coord in normalized_coords]
g.write(f"{class_idx} {' '.join(formatted_coords)}\n")
for phase in ["train", "val"]:
image_dir = dota_root_path / "images" / phase
orig_label_dir = dota_root_path / "labels" / f"{phase}_original"
save_dir = dota_root_path / "labels" / phase
save_dir.mkdir(parents=True, exist_ok=True)
image_paths = list(image_dir.iterdir())
for image_path in TQDM(image_paths, desc=f"Processing {phase} images"):
if image_path.suffix != ".png":
continue
image_name_without_ext = image_path.stem
img = cv2.imread(str(image_path))
h, w = img.shape[:2]
convert_label(image_name_without_ext, w, h, orig_label_dir, save_dir)
def min_index(arr1, arr2):
"""
Find a pair of indexes with the shortest distance between two arrays of 2D points.
Args:
arr1 (np.ndarray): A NumPy array of shape (N, 2) representing N 2D points.
arr2 (np.ndarray): A NumPy array of shape (M, 2) representing M 2D points.
Returns:
(tuple): A tuple containing the indexes of the points with the shortest distance in arr1 and arr2 respectively.
"""
dis = ((arr1[:, None, :] - arr2[None, :, :]) ** 2).sum(-1)
return np.unravel_index(np.argmin(dis, axis=None), dis.shape)
def merge_multi_segment(segments):
"""
Merge multiple segments into one list by connecting the coordinates with the minimum distance between each segment.
This function connects these coordinates with a thin line to merge all segments into one.
Args:
segments (List[List]): Original segmentations in COCO's JSON file.
Each element is a list of coordinates, like [segmentation1, segmentation2,...].
Returns:
s (List[np.ndarray]): A list of connected segments represented as NumPy arrays.
"""
s = []
segments = [np.array(i).reshape(-1, 2) for i in segments]
idx_list = [[] for _ in range(len(segments))]
# Record the indexes with min distance between each segment
for i in range(1, len(segments)):
idx1, idx2 = min_index(segments[i - 1], segments[i])
idx_list[i - 1].append(idx1)
idx_list[i].append(idx2)
# Use two round to connect all the segments
for k in range(2):
# Forward connection
if k == 0:
for i, idx in enumerate(idx_list):
# Middle segments have two indexes, reverse the index of middle segments
if len(idx) == 2 and idx[0] > idx[1]:
idx = idx[::-1]
segments[i] = segments[i][::-1, :]
segments[i] = np.roll(segments[i], -idx[0], axis=0)
segments[i] = np.concatenate([segments[i], segments[i][:1]])
# Deal with the first segment and the last one
if i in [0, len(idx_list) - 1]:
s.append(segments[i])
else:
idx = [0, idx[1] - idx[0]]
s.append(segments[i][idx[0] : idx[1] + 1])
else:
for i in range(len(idx_list) - 1, -1, -1):
if i not in [0, len(idx_list) - 1]:
idx = idx_list[i]
nidx = abs(idx[1] - idx[0])
s.append(segments[i][nidx:])
return s
def yolo_bbox2segment(im_dir, save_dir=None, sam_model="sam_b.pt"):
"""
Converts existing object detection dataset (bounding boxes) to segmentation dataset or oriented bounding box (OBB)
in YOLO format. Generates segmentation data using SAM auto-annotator as needed.
Args:
im_dir (str | Path): Path to image directory to convert.
save_dir (str | Path): Path to save the generated labels, labels will be saved
into `labels-segment` in the same directory level of `im_dir` if save_dir is None. Default: None.
sam_model (str): Segmentation model to use for intermediate segmentation data; optional.
Notes:
The input directory structure assumed for dataset:
- im_dir
├─ 001.jpg
├─ ..
└─ NNN.jpg
- labels
├─ 001.txt
├─ ..
└─ NNN.txt
"""
from ultralytics.data import YOLODataset
from ultralytics.utils.ops import xywh2xyxy
from ultralytics.utils import LOGGER
from ultralytics import SAM
from tqdm import tqdm
# NOTE: add placeholder to pass class index check
dataset = YOLODataset(im_dir, data=dict(names=list(range(1000))))
if len(dataset.labels[0]["segments"]) > 0: # if it's segment data
LOGGER.info("Segmentation labels detected, no need to generate new ones!")
return
LOGGER.info("Detection labels detected, generating segment labels by SAM model!")
sam_model = SAM(sam_model)
for l in tqdm(dataset.labels, total=len(dataset.labels), desc="Generating segment labels"):
h, w = l["shape"]
boxes = l["bboxes"]
if len(boxes) == 0: # skip empty labels
continue
boxes[:, [0, 2]] *= w
boxes[:, [1, 3]] *= h
im = cv2.imread(l["im_file"])
sam_results = sam_model(im, bboxes=xywh2xyxy(boxes), verbose=False, save=False)
l["segments"] = sam_results[0].masks.xyn
save_dir = Path(save_dir) if save_dir else Path(im_dir).parent / "labels-segment"
save_dir.mkdir(parents=True, exist_ok=True)
for l in dataset.labels:
texts = []
lb_name = Path(l["im_file"]).with_suffix(".txt").name
txt_file = save_dir / lb_name
cls = l["cls"]
for i, s in enumerate(l["segments"]):
line = (int(cls[i]), *s.reshape(-1))
texts.append(("%g " * len(line)).rstrip() % line)
if texts:
with open(txt_file, "a") as f:
f.writelines(text + "\n" for text in texts)
LOGGER.info(f"Generated segment labels saved in {save_dir}")

383
ultralytics/data/dataset.py
View File

@ -1,383 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import contextlib
from itertools import repeat
from multiprocessing.pool import ThreadPool
from pathlib import Path
import cv2
import numpy as np
import torch
import torchvision
from PIL import Image
from ultralytics.utils import LOCAL_RANK, NUM_THREADS, TQDM, colorstr, is_dir_writeable
from ultralytics.utils.ops import resample_segments
from .augment import Compose, Format, Instances, LetterBox, classify_augmentations, classify_transforms, v8_transforms
from .base import BaseDataset
from .utils import HELP_URL, LOGGER, get_hash, img2label_paths, verify_image, verify_image_label
# Ultralytics dataset *.cache version, >= 1.0.0 for YOLOv8
DATASET_CACHE_VERSION = "1.0.3"
class YOLODataset(BaseDataset):
"""
Dataset class for loading object detection and/or segmentation labels in YOLO format.
Args:
data (dict, optional): A dataset YAML dictionary. Defaults to None.
task (str): An explicit arg to point current task, Defaults to 'detect'.
Returns:
(torch.utils.data.Dataset): A PyTorch dataset object that can be used for training an object detection model.
"""
def __init__(self, *args, data=None, task="detect", **kwargs):
"""Initializes the YOLODataset with optional configurations for segments and keypoints."""
self.use_segments = task == "segment"
self.use_keypoints = task == "pose"
self.use_obb = task == "obb"
self.data = data
assert not (self.use_segments and self.use_keypoints), "Can not use both segments and keypoints."
super().__init__(*args, **kwargs)
def cache_labels(self, path=Path("./labels.cache")):
"""
Cache dataset labels, check images and read shapes.
Args:
path (Path): Path where to save the cache file. Default is Path('./labels.cache').
Returns:
(dict): labels.
"""
x = {"labels": []}
nm, nf, ne, nc, msgs = 0, 0, 0, 0, [] # number missing, found, empty, corrupt, messages
desc = f"{self.prefix}Scanning {path.parent / path.stem}..."
total = len(self.im_files)
nkpt, ndim = self.data.get("kpt_shape", (0, 0))
if self.use_keypoints and (nkpt <= 0 or ndim not in (2, 3)):
raise ValueError(
"'kpt_shape' in data.yaml missing or incorrect. Should be a list with [number of "
"keypoints, number of dims (2 for x,y or 3 for x,y,visible)], i.e. 'kpt_shape: [17, 3]'"
)
with ThreadPool(NUM_THREADS) as pool:
results = pool.imap(
func=verify_image_label,
iterable=zip(
self.im_files,
self.label_files,
repeat(self.prefix),
repeat(self.use_keypoints),
repeat(len(self.data["names"])),
repeat(nkpt),
repeat(ndim),
),
)
pbar = TQDM(results, desc=desc, total=total)
for im_file, lb, shape, segments, keypoint, nm_f, nf_f, ne_f, nc_f, msg in pbar:
nm += nm_f
nf += nf_f
ne += ne_f
nc += nc_f
if im_file:
x["labels"].append(
dict(
im_file=im_file,
shape=shape,
cls=lb[:, 0:1], # n, 1
bboxes=lb[:, 1:], # n, 4
segments=segments,
keypoints=keypoint,
normalized=True,
bbox_format="xywh",
)
)
if msg:
msgs.append(msg)
pbar.desc = f"{desc} {nf} images, {nm + ne} backgrounds, {nc} corrupt"
pbar.close()
if msgs:
LOGGER.info("\n".join(msgs))
if nf == 0:
LOGGER.warning(f"{self.prefix}WARNING ⚠️ No labels found in {path}. {HELP_URL}")
x["hash"] = get_hash(self.label_files + self.im_files)
x["results"] = nf, nm, ne, nc, len(self.im_files)
x["msgs"] = msgs # warnings
save_dataset_cache_file(self.prefix, path, x)
return x
def get_labels(self):
"""Returns dictionary of labels for YOLO training."""
self.label_files = img2label_paths(self.im_files)
cache_path = Path(self.label_files[0]).parent.with_suffix(".cache")
try:
cache, exists = load_dataset_cache_file(cache_path), True # attempt to load a *.cache file
assert cache["version"] == DATASET_CACHE_VERSION # matches current version
assert cache["hash"] == get_hash(self.label_files + self.im_files) # identical hash
except (FileNotFoundError, AssertionError, AttributeError):
cache, exists = self.cache_labels(cache_path), False # run cache ops
# Display cache
nf, nm, ne, nc, n = cache.pop("results") # found, missing, empty, corrupt, total
if exists and LOCAL_RANK in (-1, 0):
d = f"Scanning {cache_path}... {nf} images, {nm + ne} backgrounds, {nc} corrupt"
TQDM(None, desc=self.prefix + d, total=n, initial=n) # display results
if cache["msgs"]:
LOGGER.info("\n".join(cache["msgs"])) # display warnings
# Read cache
[cache.pop(k) for k in ("hash", "version", "msgs")] # remove items
labels = cache["labels"]
if not labels:
LOGGER.warning(f"WARNING ⚠️ No images found in {cache_path}, training may not work correctly. {HELP_URL}")
self.im_files = [lb["im_file"] for lb in labels] # update im_files
# Check if the dataset is all boxes or all segments
lengths = ((len(lb["cls"]), len(lb["bboxes"]), len(lb["segments"])) for lb in labels)
len_cls, len_boxes, len_segments = (sum(x) for x in zip(*lengths))
if len_segments and len_boxes != len_segments:
LOGGER.warning(
f"WARNING ⚠️ Box and segment counts should be equal, but got len(segments) = {len_segments}, "
f"len(boxes) = {len_boxes}. To resolve this only boxes will be used and all segments will be removed. "
"To avoid this please supply either a detect or segment dataset, not a detect-segment mixed dataset."
)
for lb in labels:
lb["segments"] = []
if len_cls == 0:
LOGGER.warning(f"WARNING ⚠️ No labels found in {cache_path}, training may not work correctly. {HELP_URL}")
return labels
def build_transforms(self, hyp=None):
"""Builds and appends transforms to the list."""
if self.augment:
hyp.mosaic = hyp.mosaic if self.augment and not self.rect else 0.0
hyp.mixup = hyp.mixup if self.augment and not self.rect else 0.0
transforms = v8_transforms(self, self.imgsz, hyp)
else:
transforms = Compose([LetterBox(new_shape=(self.imgsz, self.imgsz), scaleup=False)])
transforms.append(
Format(
bbox_format="xywh",
normalize=True,
return_mask=self.use_segments,
return_keypoint=self.use_keypoints,
return_obb=self.use_obb,
batch_idx=True,
mask_ratio=hyp.mask_ratio,
mask_overlap=hyp.overlap_mask,
bgr=hyp.bgr if self.augment else 0.0, # only affect training.
)
)
return transforms
def close_mosaic(self, hyp):
"""Sets mosaic, copy_paste and mixup options to 0.0 and builds transformations."""
hyp.mosaic = 0.0 # set mosaic ratio=0.0
hyp.copy_paste = 0.0 # keep the same behavior as previous v8 close-mosaic
hyp.mixup = 0.0 # keep the same behavior as previous v8 close-mosaic
self.transforms = self.build_transforms(hyp)
def update_labels_info(self, label):
"""
Custom your label format here.
Note:
cls is not with bboxes now, classification and semantic segmentation need an independent cls label
Can also support classification and semantic segmentation by adding or removing dict keys there.
"""
bboxes = label.pop("bboxes")
segments = label.pop("segments", [])
keypoints = label.pop("keypoints", None)
bbox_format = label.pop("bbox_format")
normalized = label.pop("normalized")
# NOTE: do NOT resample oriented boxes
segment_resamples = 100 if self.use_obb else 1000
if len(segments) > 0:
# list[np.array(1000, 2)] * num_samples
# (N, 1000, 2)
segments = np.stack(resample_segments(segments, n=segment_resamples), axis=0)
else:
segments = np.zeros((0, segment_resamples, 2), dtype=np.float32)
label["instances"] = Instances(bboxes, segments, keypoints, bbox_format=bbox_format, normalized=normalized)
return label
@staticmethod
def collate_fn(batch):
"""Collates data samples into batches."""
new_batch = {}
keys = batch[0].keys()
values = list(zip(*[list(b.values()) for b in batch]))
for i, k in enumerate(keys):
value = values[i]
if k == "img":
value = torch.stack(value, 0)
if k in ["masks", "keypoints", "bboxes", "cls", "segments", "obb"]:
value = torch.cat(value, 0)
new_batch[k] = value
new_batch["batch_idx"] = list(new_batch["batch_idx"])
for i in range(len(new_batch["batch_idx"])):
new_batch["batch_idx"][i] += i # add target image index for build_targets()
new_batch["batch_idx"] = torch.cat(new_batch["batch_idx"], 0)
return new_batch
# Classification dataloaders -------------------------------------------------------------------------------------------
class ClassificationDataset(torchvision.datasets.ImageFolder):
"""
Extends torchvision ImageFolder to support YOLO classification tasks, offering functionalities like image
augmentation, caching, and verification. It's designed to efficiently handle large datasets for training deep
learning models, with optional image transformations and caching mechanisms to speed up training.
This class allows for augmentations using both torchvision and Albumentations libraries, and supports caching images
in RAM or on disk to reduce IO overhead during training. Additionally, it implements a robust verification process
to ensure data integrity and consistency.
Attributes:
cache_ram (bool): Indicates if caching in RAM is enabled.
cache_disk (bool): Indicates if caching on disk is enabled.
samples (list): A list of tuples, each containing the path to an image, its class index, path to its .npy cache
file (if caching on disk), and optionally the loaded image array (if caching in RAM).
torch_transforms (callable): PyTorch transforms to be applied to the images.
"""
def __init__(self, root, args, augment=False, prefix=""):
"""
Initialize YOLO object with root, image size, augmentations, and cache settings.
Args:
root (str): Path to the dataset directory where images are stored in a class-specific folder structure.
args (Namespace): Configuration containing dataset-related settings such as image size, augmentation
parameters, and cache settings. It includes attributes like `imgsz` (image size), `fraction` (fraction
of data to use), `scale`, `fliplr`, `flipud`, `cache` (disk or RAM caching for faster training),
`auto_augment`, `hsv_h`, `hsv_s`, `hsv_v`, and `crop_fraction`.
augment (bool, optional): Whether to apply augmentations to the dataset. Default is False.
prefix (str, optional): Prefix for logging and cache filenames, aiding in dataset identification and
debugging. Default is an empty string.
"""
super().__init__(root=root)
if augment and args.fraction < 1.0: # reduce training fraction
self.samples = self.samples[: round(len(self.samples) * args.fraction)]
self.prefix = colorstr(f"{prefix}: ") if prefix else ""
self.cache_ram = args.cache is True or args.cache == "ram" # cache images into RAM
self.cache_disk = args.cache == "disk" # cache images on hard drive as uncompressed *.npy files
self.samples = self.verify_images() # filter out bad images
self.samples = [list(x) + [Path(x[0]).with_suffix(".npy"), None] for x in self.samples] # file, index, npy, im
scale = (1.0 - args.scale, 1.0) # (0.08, 1.0)
self.torch_transforms = (
classify_augmentations(
size=args.imgsz,
scale=scale,
hflip=args.fliplr,
vflip=args.flipud,
erasing=args.erasing,
auto_augment=args.auto_augment,
hsv_h=args.hsv_h,
hsv_s=args.hsv_s,
hsv_v=args.hsv_v,
)
if augment
else classify_transforms(size=args.imgsz, crop_fraction=args.crop_fraction)
)
def __getitem__(self, i):
"""Returns subset of data and targets corresponding to given indices."""
f, j, fn, im = self.samples[i] # filename, index, filename.with_suffix('.npy'), image
if self.cache_ram and im is None:
im = self.samples[i][3] = cv2.imread(f)
elif self.cache_disk:
if not fn.exists(): # load npy
np.save(fn.as_posix(), cv2.imread(f), allow_pickle=False)
im = np.load(fn)
else: # read image
im = cv2.imread(f) # BGR
# Convert NumPy array to PIL image
im = Image.fromarray(cv2.cvtColor(im, cv2.COLOR_BGR2RGB))
sample = self.torch_transforms(im)
return {"img": sample, "cls": j}
def __len__(self) -> int:
"""Return the total number of samples in the dataset."""
return len(self.samples)
def verify_images(self):
"""Verify all images in dataset."""
desc = f"{self.prefix}Scanning {self.root}..."
path = Path(self.root).with_suffix(".cache") # *.cache file path
with contextlib.suppress(FileNotFoundError, AssertionError, AttributeError):
cache = load_dataset_cache_file(path) # attempt to load a *.cache file
assert cache["version"] == DATASET_CACHE_VERSION # matches current version
assert cache["hash"] == get_hash([x[0] for x in self.samples]) # identical hash
nf, nc, n, samples = cache.pop("results") # found, missing, empty, corrupt, total
if LOCAL_RANK in (-1, 0):
d = f"{desc} {nf} images, {nc} corrupt"
TQDM(None, desc=d, total=n, initial=n)
if cache["msgs"]:
LOGGER.info("\n".join(cache["msgs"])) # display warnings
return samples
# Run scan if *.cache retrieval failed
nf, nc, msgs, samples, x = 0, 0, [], [], {}
with ThreadPool(NUM_THREADS) as pool:
results = pool.imap(func=verify_image, iterable=zip(self.samples, repeat(self.prefix)))
pbar = TQDM(results, desc=desc, total=len(self.samples))
for sample, nf_f, nc_f, msg in pbar:
if nf_f:
samples.append(sample)
if msg:
msgs.append(msg)
nf += nf_f
nc += nc_f
pbar.desc = f"{desc} {nf} images, {nc} corrupt"
pbar.close()
if msgs:
LOGGER.info("\n".join(msgs))
x["hash"] = get_hash([x[0] for x in self.samples])
x["results"] = nf, nc, len(samples), samples
x["msgs"] = msgs # warnings
save_dataset_cache_file(self.prefix, path, x)
return samples
def load_dataset_cache_file(path):
"""Load an Ultralytics *.cache dictionary from path."""
import gc
gc.disable() # reduce pickle load time https://github.com/ultralytics/ultralytics/pull/1585
cache = np.load(str(path), allow_pickle=True).item() # load dict
gc.enable()
return cache
def save_dataset_cache_file(prefix, path, x):
"""Save an Ultralytics dataset *.cache dictionary x to path."""
x["version"] = DATASET_CACHE_VERSION # add cache version
if is_dir_writeable(path.parent):
if path.exists():
path.unlink() # remove *.cache file if exists
np.save(str(path), x) # save cache for next time
path.with_suffix(".cache.npy").rename(path) # remove .npy suffix
LOGGER.info(f"{prefix}New cache created: {path}")
else:
LOGGER.warning(f"{prefix}WARNING ⚠️ Cache directory {path.parent} is not writeable, cache not saved.")
# TODO: support semantic segmentation
class SemanticDataset(BaseDataset):
"""
Semantic Segmentation Dataset.
This class is responsible for handling datasets used for semantic segmentation tasks. It inherits functionalities
from the BaseDataset class.
Note:
This class is currently a placeholder and needs to be populated with methods and attributes for supporting
semantic segmentation tasks.
"""
def __init__(self):
"""Initialize a SemanticDataset object."""
super().__init__()

5
ultralytics/data/explorer/__init__.py
View File

@ -1,5 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
from .utils import plot_query_result
__all__ = ["plot_query_result"]

472
ultralytics/data/explorer/explorer.py
View File

@ -1,472 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
from io import BytesIO
from pathlib import Path
from typing import Any, List, Tuple, Union
import cv2
import numpy as np
import torch
from PIL import Image
from matplotlib import pyplot as plt
from pandas import DataFrame
from tqdm import tqdm
from ultralytics.data.augment import Format
from ultralytics.data.dataset import YOLODataset
from ultralytics.data.utils import check_det_dataset
from ultralytics.models.yolo.model import YOLO
from ultralytics.utils import LOGGER, IterableSimpleNamespace, checks, USER_CONFIG_DIR
from .utils import get_sim_index_schema, get_table_schema, plot_query_result, prompt_sql_query, sanitize_batch
class ExplorerDataset(YOLODataset):
def __init__(self, *args, data: dict = None, **kwargs) -> None:
super().__init__(*args, data=data, **kwargs)
def load_image(self, i: int) -> Union[Tuple[np.ndarray, Tuple[int, int], Tuple[int, int]], Tuple[None, None, None]]:
"""Loads 1 image from dataset index 'i' without any resize ops."""
im, f, fn = self.ims[i], self.im_files[i], self.npy_files[i]
if im is None: # not cached in RAM
if fn.exists(): # load npy
im = np.load(fn)
else: # read image
im = cv2.imread(f) # BGR
if im is None:
raise FileNotFoundError(f"Image Not Found {f}")
h0, w0 = im.shape[:2] # orig hw
return im, (h0, w0), im.shape[:2]
return self.ims[i], self.im_hw0[i], self.im_hw[i]
def build_transforms(self, hyp: IterableSimpleNamespace = None):
"""Creates transforms for dataset images without resizing."""
return Format(
bbox_format="xyxy",
normalize=False,
return_mask=self.use_segments,
return_keypoint=self.use_keypoints,
batch_idx=True,
mask_ratio=hyp.mask_ratio,
mask_overlap=hyp.overlap_mask,
)
class Explorer:
def __init__(
self,
data: Union[str, Path] = "coco128.yaml",
model: str = "yolov8n.pt",
uri: str = USER_CONFIG_DIR / "explorer",
) -> None:
# Note duckdb==0.10.0 bug https://github.com/ultralytics/ultralytics/pull/8181
checks.check_requirements(["lancedb>=0.4.3", "duckdb<=0.9.2"])
import lancedb
self.connection = lancedb.connect(uri)
self.table_name = Path(data).name.lower() + "_" + model.lower()
self.sim_idx_base_name = (
f"{self.table_name}_sim_idx".lower()
) # Use this name and append thres and top_k to reuse the table
self.model = YOLO(model)
self.data = data # None
self.choice_set = None
self.table = None
self.progress = 0
def create_embeddings_table(self, force: bool = False, split: str = "train") -> None:
"""
Create LanceDB table containing the embeddings of the images in the dataset. The table will be reused if it
already exists. Pass force=True to overwrite the existing table.
Args:
force (bool): Whether to overwrite the existing table or not. Defaults to False.
split (str): Split of the dataset to use. Defaults to 'train'.
Example:
```python
exp = Explorer()
exp.create_embeddings_table()
```
"""
if self.table is not None and not force:
LOGGER.info("Table already exists. Reusing it. Pass force=True to overwrite it.")
return
if self.table_name in self.connection.table_names() and not force:
LOGGER.info(f"Table {self.table_name} already exists. Reusing it. Pass force=True to overwrite it.")
self.table = self.connection.open_table(self.table_name)
self.progress = 1
return
if self.data is None:
raise ValueError("Data must be provided to create embeddings table")
data_info = check_det_dataset(self.data)
if split not in data_info:
raise ValueError(
f"Split {split} is not found in the dataset. Available keys in the dataset are {list(data_info.keys())}"
)
choice_set = data_info[split]
choice_set = choice_set if isinstance(choice_set, list) else [choice_set]
self.choice_set = choice_set
dataset = ExplorerDataset(img_path=choice_set, data=data_info, augment=False, cache=False, task=self.model.task)
# Create the table schema
batch = dataset[0]
vector_size = self.model.embed(batch["im_file"], verbose=False)[0].shape[0]
table = self.connection.create_table(self.table_name, schema=get_table_schema(vector_size), mode="overwrite")
table.add(
self._yield_batches(
dataset,
data_info,
self.model,
exclude_keys=["img", "ratio_pad", "resized_shape", "ori_shape", "batch_idx"],
)
)
self.table = table
def _yield_batches(self, dataset: ExplorerDataset, data_info: dict, model: YOLO, exclude_keys: List[str]):
"""Generates batches of data for embedding, excluding specified keys."""
for i in tqdm(range(len(dataset))):
self.progress = float(i + 1) / len(dataset)
batch = dataset[i]
for k in exclude_keys:
batch.pop(k, None)
batch = sanitize_batch(batch, data_info)
batch["vector"] = model.embed(batch["im_file"], verbose=False)[0].detach().tolist()
yield [batch]
def query(
self, imgs: Union[str, np.ndarray, List[str], List[np.ndarray]] = None, limit: int = 25
) -> Any: # pyarrow.Table
"""
Query the table for similar images. Accepts a single image or a list of images.
Args:
imgs (str or list): Path to the image or a list of paths to the images.
limit (int): Number of results to return.
Returns:
(pyarrow.Table): An arrow table containing the results. Supports converting to:
- pandas dataframe: `result.to_pandas()`
- dict of lists: `result.to_pydict()`
Example:
```python
exp = Explorer()
exp.create_embeddings_table()
similar = exp.query(img='https://ultralytics.com/images/zidane.jpg')
```
"""
if self.table is None:
raise ValueError("Table is not created. Please create the table first.")
if isinstance(imgs, str):
imgs = [imgs]
assert isinstance(imgs, list), f"img must be a string or a list of strings. Got {type(imgs)}"
embeds = self.model.embed(imgs)
# Get avg if multiple images are passed (len > 1)
embeds = torch.mean(torch.stack(embeds), 0).cpu().numpy() if len(embeds) > 1 else embeds[0].cpu().numpy()
return self.table.search(embeds).limit(limit).to_arrow()
def sql_query(
self, query: str, return_type: str = "pandas"
) -> Union[DataFrame, Any, None]: # pandas.dataframe or pyarrow.Table
"""
Run a SQL-Like query on the table. Utilizes LanceDB predicate pushdown.
Args:
query (str): SQL query to run.
return_type (str): Type of the result to return. Can be either 'pandas' or 'arrow'. Defaults to 'pandas'.
Returns:
(pyarrow.Table): An arrow table containing the results.
Example:
```python
exp = Explorer()
exp.create_embeddings_table()
query = "SELECT * FROM 'table' WHERE labels LIKE '%person%'"
result = exp.sql_query(query)
```
"""
assert return_type in {
"pandas",
"arrow",
}, f"Return type should be either `pandas` or `arrow`, but got {return_type}"
import duckdb
if self.table is None:
raise ValueError("Table is not created. Please create the table first.")
# Note: using filter pushdown would be a better long term solution. Temporarily using duckdb for this.
table = self.table.to_arrow() # noqa NOTE: Don't comment this. This line is used by DuckDB
if not query.startswith("SELECT") and not query.startswith("WHERE"):
raise ValueError(
f"Query must start with SELECT or WHERE. You can either pass the entire query or just the WHERE clause. found {query}"
)
if query.startswith("WHERE"):
query = f"SELECT * FROM 'table' {query}"
LOGGER.info(f"Running query: {query}")
rs = duckdb.sql(query)
if return_type == "arrow":
return rs.arrow()
elif return_type == "pandas":
return rs.df()
def plot_sql_query(self, query: str, labels: bool = True) -> Image.Image:
"""
Plot the results of a SQL-Like query on the table.
Args:
query (str): SQL query to run.
labels (bool): Whether to plot the labels or not.
Returns:
(PIL.Image): Image containing the plot.
Example:
```python
exp = Explorer()
exp.create_embeddings_table()
query = "SELECT * FROM 'table' WHERE labels LIKE '%person%'"
result = exp.plot_sql_query(query)
```
"""
result = self.sql_query(query, return_type="arrow")
if len(result) == 0:
LOGGER.info("No results found.")
return None
img = plot_query_result(result, plot_labels=labels)
return Image.fromarray(img)
def get_similar(
self,
img: Union[str, np.ndarray, List[str], List[np.ndarray]] = None,
idx: Union[int, List[int]] = None,
limit: int = 25,
return_type: str = "pandas",
) -> Union[DataFrame, Any]: # pandas.dataframe or pyarrow.Table
"""
Query the table for similar images. Accepts a single image or a list of images.
Args:
img (str or list): Path to the image or a list of paths to the images.
idx (int or list): Index of the image in the table or a list of indexes.
limit (int): Number of results to return. Defaults to 25.
return_type (str): Type of the result to return. Can be either 'pandas' or 'arrow'. Defaults to 'pandas'.
Returns:
(pandas.DataFrame): A dataframe containing the results.
Example:
```python
exp = Explorer()
exp.create_embeddings_table()
similar = exp.get_similar(img='https://ultralytics.com/images/zidane.jpg')
```
"""
assert return_type in {
"pandas",
"arrow",
}, f"Return type should be either `pandas` or `arrow`, but got {return_type}"
img = self._check_imgs_or_idxs(img, idx)
similar = self.query(img, limit=limit)
if return_type == "arrow":
return similar
elif return_type == "pandas":
return similar.to_pandas()
def plot_similar(
self,
img: Union[str, np.ndarray, List[str], List[np.ndarray]] = None,
idx: Union[int, List[int]] = None,
limit: int = 25,
labels: bool = True,
) -> Image.Image:
"""
Plot the similar images. Accepts images or indexes.
Args:
img (str or list): Path to the image or a list of paths to the images.
idx (int or list): Index of the image in the table or a list of indexes.
labels (bool): Whether to plot the labels or not.
limit (int): Number of results to return. Defaults to 25.
Returns:
(PIL.Image): Image containing the plot.
Example:
```python
exp = Explorer()
exp.create_embeddings_table()
similar = exp.plot_similar(img='https://ultralytics.com/images/zidane.jpg')
```
"""
similar = self.get_similar(img, idx, limit, return_type="arrow")
if len(similar) == 0:
LOGGER.info("No results found.")
return None
img = plot_query_result(similar, plot_labels=labels)
return Image.fromarray(img)
def similarity_index(self, max_dist: float = 0.2, top_k: float = None, force: bool = False) -> DataFrame:
"""
Calculate the similarity index of all the images in the table. Here, the index will contain the data points that
are max_dist or closer to the image in the embedding space at a given index.
Args:
max_dist (float): maximum L2 distance between the embeddings to consider. Defaults to 0.2.
top_k (float): Percentage of the closest data points to consider when counting. Used to apply limit when running
vector search. Defaults: None.
force (bool): Whether to overwrite the existing similarity index or not. Defaults to True.
Returns:
(pandas.DataFrame): A dataframe containing the similarity index. Each row corresponds to an image, and columns
include indices of similar images and their respective distances.
Example:
```python
exp = Explorer()
exp.create_embeddings_table()
sim_idx = exp.similarity_index()
```
"""
if self.table is None:
raise ValueError("Table is not created. Please create the table first.")
sim_idx_table_name = f"{self.sim_idx_base_name}_thres_{max_dist}_top_{top_k}".lower()
if sim_idx_table_name in self.connection.table_names() and not force:
LOGGER.info("Similarity matrix already exists. Reusing it. Pass force=True to overwrite it.")
return self.connection.open_table(sim_idx_table_name).to_pandas()
if top_k and not (1.0 >= top_k >= 0.0):
raise ValueError(f"top_k must be between 0.0 and 1.0. Got {top_k}")
if max_dist < 0.0:
raise ValueError(f"max_dist must be greater than 0. Got {max_dist}")
top_k = int(top_k * len(self.table)) if top_k else len(self.table)
top_k = max(top_k, 1)
features = self.table.to_lance().to_table(columns=["vector", "im_file"]).to_pydict()
im_files = features["im_file"]
embeddings = features["vector"]
sim_table = self.connection.create_table(sim_idx_table_name, schema=get_sim_index_schema(), mode="overwrite")
def _yield_sim_idx():
"""Generates a dataframe with similarity indices and distances for images."""
for i in tqdm(range(len(embeddings))):
sim_idx = self.table.search(embeddings[i]).limit(top_k).to_pandas().query(f"_distance <= {max_dist}")
yield [
{
"idx": i,
"im_file": im_files[i],
"count": len(sim_idx),
"sim_im_files": sim_idx["im_file"].tolist(),
}
]
sim_table.add(_yield_sim_idx())
self.sim_index = sim_table
return sim_table.to_pandas()
def plot_similarity_index(self, max_dist: float = 0.2, top_k: float = None, force: bool = False) -> Image:
"""
Plot the similarity index of all the images in the table. Here, the index will contain the data points that are
max_dist or closer to the image in the embedding space at a given index.
Args:
max_dist (float): maximum L2 distance between the embeddings to consider. Defaults to 0.2.
top_k (float): Percentage of closest data points to consider when counting. Used to apply limit when
running vector search. Defaults to 0.01.
force (bool): Whether to overwrite the existing similarity index or not. Defaults to True.
Returns:
(PIL.Image): Image containing the plot.
Example:
```python
exp = Explorer()
exp.create_embeddings_table()
similarity_idx_plot = exp.plot_similarity_index()
similarity_idx_plot.show() # view image preview
similarity_idx_plot.save('path/to/save/similarity_index_plot.png') # save contents to file
```
"""
sim_idx = self.similarity_index(max_dist=max_dist, top_k=top_k, force=force)
sim_count = sim_idx["count"].tolist()
sim_count = np.array(sim_count)
indices = np.arange(len(sim_count))
# Create the bar plot
plt.bar(indices, sim_count)
# Customize the plot (optional)
plt.xlabel("data idx")
plt.ylabel("Count")
plt.title("Similarity Count")
buffer = BytesIO()
plt.savefig(buffer, format="png")
buffer.seek(0)
# Use Pillow to open the image from the buffer
return Image.fromarray(np.array(Image.open(buffer)))
def _check_imgs_or_idxs(
self, img: Union[str, np.ndarray, List[str], List[np.ndarray], None], idx: Union[None, int, List[int]]
) -> List[np.ndarray]:
if img is None and idx is None:
raise ValueError("Either img or idx must be provided.")
if img is not None and idx is not None:
raise ValueError("Only one of img or idx must be provided.")
if idx is not None:
idx = idx if isinstance(idx, list) else [idx]
img = self.table.to_lance().take(idx, columns=["im_file"]).to_pydict()["im_file"]
return img if isinstance(img, list) else [img]
def ask_ai(self, query):
"""
Ask AI a question.
Args:
query (str): Question to ask.
Returns:
(pandas.DataFrame): A dataframe containing filtered results to the SQL query.
Example:
```python
exp = Explorer()
exp.create_embeddings_table()
answer = exp.ask_ai('Show images with 1 person and 2 dogs')
```
"""
result = prompt_sql_query(query)
try:
df = self.sql_query(result)
except Exception as e:
LOGGER.error("AI generated query is not valid. Please try again with a different prompt")
LOGGER.error(e)
return None
return df
def visualize(self, result):
"""
Visualize the results of a query. TODO.
Args:
result (pyarrow.Table): Table containing the results of a query.
"""
pass
def generate_report(self, result):
"""
Generate a report of the dataset.
TODO
"""
pass

1
ultralytics/data/explorer/gui/__init__.py
View File

@ -1 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license

268
ultralytics/data/explorer/gui/dash.py
View File

@ -1,268 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import time
from threading import Thread
import pandas as pd
from ultralytics import Explorer
from ultralytics.utils import ROOT, SETTINGS
from ultralytics.utils.checks import check_requirements
check_requirements(("streamlit>=1.29.0", "streamlit-select>=0.3"))
import streamlit as st
from streamlit_select import image_select
def _get_explorer():
"""Initializes and returns an instance of the Explorer class."""
exp = Explorer(data=st.session_state.get("dataset"), model=st.session_state.get("model"))
thread = Thread(
target=exp.create_embeddings_table, kwargs={"force": st.session_state.get("force_recreate_embeddings")}
)
thread.start()
progress_bar = st.progress(0, text="Creating embeddings table...")
while exp.progress < 1:
time.sleep(0.1)
progress_bar.progress(exp.progress, text=f"Progress: {exp.progress * 100}%")
thread.join()
st.session_state["explorer"] = exp
progress_bar.empty()
def init_explorer_form():
"""Initializes an Explorer instance and creates embeddings table with progress tracking."""
datasets = ROOT / "cfg" / "datasets"
ds = [d.name for d in datasets.glob("*.yaml")]
models = [
"yolov8n.pt",
"yolov8s.pt",
"yolov8m.pt",
"yolov8l.pt",
"yolov8x.pt",
"yolov8n-seg.pt",
"yolov8s-seg.pt",
"yolov8m-seg.pt",
"yolov8l-seg.pt",
"yolov8x-seg.pt",
"yolov8n-pose.pt",
"yolov8s-pose.pt",
"yolov8m-pose.pt",
"yolov8l-pose.pt",
"yolov8x-pose.pt",
]
with st.form(key="explorer_init_form"):
col1, col2 = st.columns(2)
with col1:
st.selectbox("Select dataset", ds, key="dataset", index=ds.index("coco128.yaml"))
with col2:
st.selectbox("Select model", models, key="model")
st.checkbox("Force recreate embeddings", key="force_recreate_embeddings")
st.form_submit_button("Explore", on_click=_get_explorer)
def query_form():
"""Sets up a form in Streamlit to initialize Explorer with dataset and model selection."""
with st.form("query_form"):
col1, col2 = st.columns([0.8, 0.2])
with col1:
st.text_input(
"Query",
"WHERE labels LIKE '%person%' AND labels LIKE '%dog%'",
label_visibility="collapsed",
key="query",
)
with col2:
st.form_submit_button("Query", on_click=run_sql_query)
def ai_query_form():
"""Sets up a Streamlit form for user input to initialize Explorer with dataset and model selection."""
with st.form("ai_query_form"):
col1, col2 = st.columns([0.8, 0.2])
with col1:
st.text_input("Query", "Show images with 1 person and 1 dog", label_visibility="collapsed", key="ai_query")
with col2:
st.form_submit_button("Ask AI", on_click=run_ai_query)
def find_similar_imgs(imgs):
"""Initializes a Streamlit form for AI-based image querying with custom input."""
exp = st.session_state["explorer"]
similar = exp.get_similar(img=imgs, limit=st.session_state.get("limit"), return_type="arrow")
paths = similar.to_pydict()["im_file"]
st.session_state["imgs"] = paths
st.session_state["res"] = similar
def similarity_form(selected_imgs):
"""Initializes a form for AI-based image querying with custom input in Streamlit."""
st.write("Similarity Search")
with st.form("similarity_form"):
subcol1, subcol2 = st.columns([1, 1])
with subcol1:
st.number_input(
"limit", min_value=None, max_value=None, value=25, label_visibility="collapsed", key="limit"
)
with subcol2:
disabled = not len(selected_imgs)
st.write("Selected: ", len(selected_imgs))
st.form_submit_button(
"Search",
disabled=disabled,
on_click=find_similar_imgs,
args=(selected_imgs,),
)
if disabled:
st.error("Select at least one image to search.")
# def persist_reset_form():
# with st.form("persist_reset"):
# col1, col2 = st.columns([1, 1])
# with col1:
# st.form_submit_button("Reset", on_click=reset)
#
# with col2:
# st.form_submit_button("Persist", on_click=update_state, args=("PERSISTING", True))
def run_sql_query():
"""Executes an SQL query and returns the results."""
st.session_state["error"] = None
query = st.session_state.get("query")
if query.rstrip().lstrip():
exp = st.session_state["explorer"]
res = exp.sql_query(query, return_type="arrow")
st.session_state["imgs"] = res.to_pydict()["im_file"]
st.session_state["res"] = res
def run_ai_query():
"""Execute SQL query and update session state with query results."""
if not SETTINGS["openai_api_key"]:
st.session_state["error"] = (
'OpenAI API key not found in settings. Please run yolo settings openai_api_key="..."'
)
return
st.session_state["error"] = None
query = st.session_state.get("ai_query")
if query.rstrip().lstrip():
exp = st.session_state["explorer"]
res = exp.ask_ai(query)
if not isinstance(res, pd.DataFrame) or res.empty:
st.session_state["error"] = "No results found using AI generated query. Try another query or rerun it."
return
st.session_state["imgs"] = res["im_file"].to_list()
st.session_state["res"] = res
def reset_explorer():
"""Resets the explorer to its initial state by clearing session variables."""
st.session_state["explorer"] = None
st.session_state["imgs"] = None
st.session_state["error"] = None
def utralytics_explorer_docs_callback():
"""Resets the explorer to its initial state by clearing session variables."""
with st.container(border=True):
st.image(
"https://raw.githubusercontent.com/ultralytics/assets/main/logo/Ultralytics_Logotype_Original.svg",
width=100,
)
st.markdown(
"<p>This demo is built using Ultralytics Explorer API. Visit <a href='https://docs.ultralytics.com/datasets/explorer/'>API docs</a> to try examples & learn more</p>",
unsafe_allow_html=True,
help=None,
)
st.link_button("Ultrlaytics Explorer API", "https://docs.ultralytics.com/datasets/explorer/")
def layout():
"""Resets explorer session variables and provides documentation with a link to API docs."""
st.set_page_config(layout="wide", initial_sidebar_state="collapsed")
st.markdown("<h1 style='text-align: center;'>Ultralytics Explorer Demo</h1>", unsafe_allow_html=True)
if st.session_state.get("explorer") is None:
init_explorer_form()
return
st.button(":arrow_backward: Select Dataset", on_click=reset_explorer)
exp = st.session_state.get("explorer")
col1, col2 = st.columns([0.75, 0.25], gap="small")
imgs = []
if st.session_state.get("error"):
st.error(st.session_state["error"])
else:
if st.session_state.get("imgs"):
imgs = st.session_state.get("imgs")
else:
imgs = exp.table.to_lance().to_table(columns=["im_file"]).to_pydict()["im_file"]
st.session_state["res"] = exp.table.to_arrow()
total_imgs, selected_imgs = len(imgs), []
with col1:
subcol1, subcol2, subcol3, subcol4, subcol5 = st.columns(5)
with subcol1:
st.write("Max Images Displayed:")
with subcol2:
num = st.number_input(
"Max Images Displayed",
min_value=0,
max_value=total_imgs,
value=min(500, total_imgs),
key="num_imgs_displayed",
label_visibility="collapsed",
)
with subcol3:
st.write("Start Index:")
with subcol4:
start_idx = st.number_input(
"Start Index",
min_value=0,
max_value=total_imgs,
value=0,
key="start_index",
label_visibility="collapsed",
)
with subcol5:
reset = st.button("Reset", use_container_width=False, key="reset")
if reset:
st.session_state["imgs"] = None
st.experimental_rerun()
query_form()
ai_query_form()
if total_imgs:
labels, boxes, masks, kpts, classes = None, None, None, None, None
task = exp.model.task
if st.session_state.get("display_labels"):
labels = st.session_state.get("res").to_pydict()["labels"][start_idx : start_idx + num]
boxes = st.session_state.get("res").to_pydict()["bboxes"][start_idx : start_idx + num]
masks = st.session_state.get("res").to_pydict()["masks"][start_idx : start_idx + num]
kpts = st.session_state.get("res").to_pydict()["keypoints"][start_idx : start_idx + num]
classes = st.session_state.get("res").to_pydict()["cls"][start_idx : start_idx + num]
imgs_displayed = imgs[start_idx : start_idx + num]
selected_imgs = image_select(
f"Total samples: {total_imgs}",
images=imgs_displayed,
use_container_width=False,
# indices=[i for i in range(num)] if select_all else None,
labels=labels,
classes=classes,
bboxes=boxes,
masks=masks if task == "segment" else None,
kpts=kpts if task == "pose" else None,
)
with col2:
similarity_form(selected_imgs)
display_labels = st.checkbox("Labels", value=False, key="display_labels")
utralytics_explorer_docs_callback()
if __name__ == "__main__":
layout()

166
ultralytics/data/explorer/utils.py
View File

@ -1,166 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import getpass
from typing import List
import cv2
import numpy as np
import pandas as pd
from ultralytics.data.augment import LetterBox
from ultralytics.utils import LOGGER as logger
from ultralytics.utils import SETTINGS
from ultralytics.utils.checks import check_requirements
from ultralytics.utils.ops import xyxy2xywh
from ultralytics.utils.plotting import plot_images
def get_table_schema(vector_size):
"""Extracts and returns the schema of a database table."""
from lancedb.pydantic import LanceModel, Vector
class Schema(LanceModel):
im_file: str
labels: List[str]
cls: List[int]
bboxes: List[List[float]]
masks: List[List[List[int]]]
keypoints: List[List[List[float]]]
vector: Vector(vector_size)
return Schema
def get_sim_index_schema():
"""Returns a LanceModel schema for a database table with specified vector size."""
from lancedb.pydantic import LanceModel
class Schema(LanceModel):
idx: int
im_file: str
count: int
sim_im_files: List[str]
return Schema
def sanitize_batch(batch, dataset_info):
"""Sanitizes input batch for inference, ensuring correct format and dimensions."""
batch["cls"] = batch["cls"].flatten().int().tolist()
box_cls_pair = sorted(zip(batch["bboxes"].tolist(), batch["cls"]), key=lambda x: x[1])
batch["bboxes"] = [box for box, _ in box_cls_pair]
batch["cls"] = [cls for _, cls in box_cls_pair]
batch["labels"] = [dataset_info["names"][i] for i in batch["cls"]]
batch["masks"] = batch["masks"].tolist() if "masks" in batch else [[[]]]
batch["keypoints"] = batch["keypoints"].tolist() if "keypoints" in batch else [[[]]]
return batch
def plot_query_result(similar_set, plot_labels=True):
"""
Plot images from the similar set.
Args:
similar_set (list): Pyarrow or pandas object containing the similar data points
plot_labels (bool): Whether to plot labels or not
"""
similar_set = (
similar_set.to_dict(orient="list") if isinstance(similar_set, pd.DataFrame) else similar_set.to_pydict()
)
empty_masks = [[[]]]
empty_boxes = [[]]
images = similar_set.get("im_file", [])
bboxes = similar_set.get("bboxes", []) if similar_set.get("bboxes") is not empty_boxes else []
masks = similar_set.get("masks") if similar_set.get("masks")[0] != empty_masks else []
kpts = similar_set.get("keypoints") if similar_set.get("keypoints")[0] != empty_masks else []
cls = similar_set.get("cls", [])
plot_size = 640
imgs, batch_idx, plot_boxes, plot_masks, plot_kpts = [], [], [], [], []
for i, imf in enumerate(images):
im = cv2.imread(imf)
im = cv2.cvtColor(im, cv2.COLOR_BGR2RGB)
h, w = im.shape[:2]
r = min(plot_size / h, plot_size / w)
imgs.append(LetterBox(plot_size, center=False)(image=im).transpose(2, 0, 1))
if plot_labels:
if len(bboxes) > i and len(bboxes[i]) > 0:
box = np.array(bboxes[i], dtype=np.float32)
box[:, [0, 2]] *= r
box[:, [1, 3]] *= r
plot_boxes.append(box)
if len(masks) > i and len(masks[i]) > 0:
mask = np.array(masks[i], dtype=np.uint8)[0]
plot_masks.append(LetterBox(plot_size, center=False)(image=mask))
if len(kpts) > i and kpts[i] is not None:
kpt = np.array(kpts[i], dtype=np.float32)
kpt[:, :, :2] *= r
plot_kpts.append(kpt)
batch_idx.append(np.ones(len(np.array(bboxes[i], dtype=np.float32))) * i)
imgs = np.stack(imgs, axis=0)
masks = np.stack(plot_masks, axis=0) if plot_masks else np.zeros(0, dtype=np.uint8)
kpts = np.concatenate(plot_kpts, axis=0) if plot_kpts else np.zeros((0, 51), dtype=np.float32)
boxes = xyxy2xywh(np.concatenate(plot_boxes, axis=0)) if plot_boxes else np.zeros(0, dtype=np.float32)
batch_idx = np.concatenate(batch_idx, axis=0)
cls = np.concatenate([np.array(c, dtype=np.int32) for c in cls], axis=0)
return plot_images(
imgs, batch_idx, cls, bboxes=boxes, masks=masks, kpts=kpts, max_subplots=len(images), save=False, threaded=False
)
def prompt_sql_query(query):
"""Plots images with optional labels from a similar data set."""
check_requirements("openai>=1.6.1")
from openai import OpenAI
if not SETTINGS["openai_api_key"]:
logger.warning("OpenAI API key not found in settings. Please enter your API key below.")
openai_api_key = getpass.getpass("OpenAI API key: ")
SETTINGS.update({"openai_api_key": openai_api_key})
openai = OpenAI(api_key=SETTINGS["openai_api_key"])
messages = [
{
"role": "system",
"content": """
You are a helpful data scientist proficient in SQL. You need to output exactly one SQL query based on
the following schema and a user request. You only need to output the format with fixed selection
statement that selects everything from "'table'", like `SELECT * from 'table'`
Schema:
im_file: string not null
labels: list<item: string> not null
child 0, item: string
cls: list<item: int64> not null
child 0, item: int64
bboxes: list<item: list<item: double>> not null
child 0, item: list<item: double>
child 0, item: double
masks: list<item: list<item: list<item: int64>>> not null
child 0, item: list<item: list<item: int64>>
child 0, item: list<item: int64>
child 0, item: int64
keypoints: list<item: list<item: list<item: double>>> not null
child 0, item: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
vector: fixed_size_list<item: float>[256] not null
child 0, item: float
Some details about the schema:
- the "labels" column contains the string values like 'person' and 'dog' for the respective objects
in each image
- the "cls" column contains the integer values on these classes that map them the labels
Example of a correct query:
request - Get all data points that contain 2 or more people and at least one dog
correct query-
SELECT * FROM 'table' WHERE ARRAY_LENGTH(cls) >= 2 AND ARRAY_LENGTH(FILTER(labels, x -> x = 'person')) >= 2 AND ARRAY_LENGTH(FILTER(labels, x -> x = 'dog')) >= 1;
""",
},
{"role": "user", "content": f"{query}"},
]
response = openai.chat.completions.create(model="gpt-3.5-turbo", messages=messages)
return response.choices[0].message.content

555
ultralytics/data/loaders.py
View File

@ -1,555 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import glob
import math
import os
import time
from dataclasses import dataclass
from pathlib import Path
from threading import Thread
from urllib.parse import urlparse
import cv2
import numpy as np
import requests
import torch
from PIL import Image
from ultralytics.data.utils import IMG_FORMATS, VID_FORMATS
from ultralytics.utils import LOGGER, is_colab, is_kaggle, ops
from ultralytics.utils.checks import check_requirements
@dataclass
class SourceTypes:
"""Class to represent various types of input sources for predictions."""
stream: bool = False
screenshot: bool = False
from_img: bool = False
tensor: bool = False
class LoadStreams:
"""
Stream Loader for various types of video streams, Supports RTSP, RTMP, HTTP, and TCP streams.
Attributes:
sources (str): The source input paths or URLs for the video streams.
vid_stride (int): Video frame-rate stride, defaults to 1.
buffer (bool): Whether to buffer input streams, defaults to False.
running (bool): Flag to indicate if the streaming thread is running.
mode (str): Set to 'stream' indicating real-time capture.
imgs (list): List of image frames for each stream.
fps (list): List of FPS for each stream.
frames (list): List of total frames for each stream.
threads (list): List of threads for each stream.
shape (list): List of shapes for each stream.
caps (list): List of cv2.VideoCapture objects for each stream.
bs (int): Batch size for processing.
Methods:
__init__: Initialize the stream loader.
update: Read stream frames in daemon thread.
close: Close stream loader and release resources.
__iter__: Returns an iterator object for the class.
__next__: Returns source paths, transformed, and original images for processing.
__len__: Return the length of the sources object.
Example:
```bash
yolo predict source='rtsp://example.com/media.mp4'
```
"""
def __init__(self, sources="file.streams", vid_stride=1, buffer=False):
"""Initialize instance variables and check for consistent input stream shapes."""
torch.backends.cudnn.benchmark = True # faster for fixed-size inference
self.buffer = buffer # buffer input streams
self.running = True # running flag for Thread
self.mode = "stream"
self.vid_stride = vid_stride # video frame-rate stride
sources = Path(sources).read_text().rsplit() if os.path.isfile(sources) else [sources]
n = len(sources)
self.bs = n
self.fps = [0] * n # frames per second
self.frames = [0] * n
self.threads = [None] * n
self.caps = [None] * n # video capture objects
self.imgs = [[] for _ in range(n)] # images
self.shape = [[] for _ in range(n)] # image shapes
self.sources = [ops.clean_str(x) for x in sources] # clean source names for later
for i, s in enumerate(sources): # index, source
# Start thread to read frames from video stream
st = f"{i + 1}/{n}: {s}... "
if urlparse(s).hostname in ("www.youtube.com", "youtube.com", "youtu.be"): # if source is YouTube video
# YouTube format i.e. 'https://www.youtube.com/watch?v=Zgi9g1ksQHc' or 'https://youtu.be/LNwODJXcvt4'
s = get_best_youtube_url(s)
s = eval(s) if s.isnumeric() else s # i.e. s = '0' local webcam
if s == 0 and (is_colab() or is_kaggle()):
raise NotImplementedError(
"'source=0' webcam not supported in Colab and Kaggle notebooks. "
"Try running 'source=0' in a local environment."
)
self.caps[i] = cv2.VideoCapture(s) # store video capture object
if not self.caps[i].isOpened():
raise ConnectionError(f"{st}Failed to open {s}")
w = int(self.caps[i].get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(self.caps[i].get(cv2.CAP_PROP_FRAME_HEIGHT))
fps = self.caps[i].get(cv2.CAP_PROP_FPS) # warning: may return 0 or nan
self.frames[i] = max(int(self.caps[i].get(cv2.CAP_PROP_FRAME_COUNT)), 0) or float(
"inf"
) # infinite stream fallback
self.fps[i] = max((fps if math.isfinite(fps) else 0) % 100, 0) or 30 # 30 FPS fallback
success, im = self.caps[i].read() # guarantee first frame
if not success or im is None:
raise ConnectionError(f"{st}Failed to read images from {s}")
self.imgs[i].append(im)
self.shape[i] = im.shape
self.threads[i] = Thread(target=self.update, args=([i, self.caps[i], s]), daemon=True)
LOGGER.info(f"{st}Success ✅ ({self.frames[i]} frames of shape {w}x{h} at {self.fps[i]:.2f} FPS)")
self.threads[i].start()
LOGGER.info("") # newline
def update(self, i, cap, stream):
"""Read stream `i` frames in daemon thread."""
n, f = 0, self.frames[i] # frame number, frame array
while self.running and cap.isOpened() and n < (f - 1):
if len(self.imgs[i]) < 30: # keep a <=30-image buffer
n += 1
cap.grab() # .read() = .grab() followed by .retrieve()
if n % self.vid_stride == 0:
success, im = cap.retrieve()
if not success:
im = np.zeros(self.shape[i], dtype=np.uint8)
LOGGER.warning("WARNING ⚠️ Video stream unresponsive, please check your IP camera connection.")
cap.open(stream) # re-open stream if signal was lost
if self.buffer:
self.imgs[i].append(im)
else:
self.imgs[i] = [im]
else:
time.sleep(0.01) # wait until the buffer is empty
def close(self):
"""Close stream loader and release resources."""
self.running = False # stop flag for Thread
for thread in self.threads:
if thread.is_alive():
thread.join(timeout=5) # Add timeout
for cap in self.caps: # Iterate through the stored VideoCapture objects
try:
cap.release() # release video capture
except Exception as e:
LOGGER.warning(f"WARNING ⚠️ Could not release VideoCapture object: {e}")
cv2.destroyAllWindows()
def __iter__(self):
"""Iterates through YOLO image feed and re-opens unresponsive streams."""
self.count = -1
return self
def __next__(self):
"""Returns source paths, transformed and original images for processing."""
self.count += 1
images = []
for i, x in enumerate(self.imgs):
# Wait until a frame is available in each buffer
while not x:
if not self.threads[i].is_alive() or cv2.waitKey(1) == ord("q"): # q to quit
self.close()
raise StopIteration
time.sleep(1 / min(self.fps))
x = self.imgs[i]
if not x:
LOGGER.warning(f"WARNING ⚠️ Waiting for stream {i}")
# Get and remove the first frame from imgs buffer
if self.buffer:
images.append(x.pop(0))
# Get the last frame, and clear the rest from the imgs buffer
else:
images.append(x.pop(-1) if x else np.zeros(self.shape[i], dtype=np.uint8))
x.clear()
return self.sources, images, [""] * self.bs
def __len__(self):
"""Return the length of the sources object."""
return self.bs # 1E12 frames = 32 streams at 30 FPS for 30 years
class LoadScreenshots:
"""
YOLOv8 screenshot dataloader.
This class manages the loading of screenshot images for processing with YOLOv8.
Suitable for use with `yolo predict source=screen`.
Attributes:
source (str): The source input indicating which screen to capture.
screen (int): The screen number to capture.
left (int): The left coordinate for screen capture area.
top (int): The top coordinate for screen capture area.
width (int): The width of the screen capture area.
height (int): The height of the screen capture area.
mode (str): Set to 'stream' indicating real-time capture.
frame (int): Counter for captured frames.
sct (mss.mss): Screen capture object from `mss` library.
bs (int): Batch size, set to 1.
monitor (dict): Monitor configuration details.
Methods:
__iter__: Returns an iterator object.
__next__: Captures the next screenshot and returns it.
"""
def __init__(self, source):
"""Source = [screen_number left top width height] (pixels)."""
check_requirements("mss")
import mss # noqa
source, *params = source.split()
self.screen, left, top, width, height = 0, None, None, None, None # default to full screen 0
if len(params) == 1:
self.screen = int(params[0])
elif len(params) == 4:
left, top, width, height = (int(x) for x in params)
elif len(params) == 5:
self.screen, left, top, width, height = (int(x) for x in params)
self.mode = "stream"
self.frame = 0
self.sct = mss.mss()
self.bs = 1
self.fps = 30
# Parse monitor shape
monitor = self.sct.monitors[self.screen]
self.top = monitor["top"] if top is None else (monitor["top"] + top)
self.left = monitor["left"] if left is None else (monitor["left"] + left)
self.width = width or monitor["width"]
self.height = height or monitor["height"]
self.monitor = {"left": self.left, "top": self.top, "width": self.width, "height": self.height}
def __iter__(self):
"""Returns an iterator of the object."""
return self
def __next__(self):
"""mss screen capture: get raw pixels from the screen as np array."""
im0 = np.asarray(self.sct.grab(self.monitor))[:, :, :3] # BGRA to BGR
s = f"screen {self.screen} (LTWH): {self.left},{self.top},{self.width},{self.height}: "
self.frame += 1
return [str(self.screen)], [im0], [s] # screen, img, string
class LoadImagesAndVideos:
"""
YOLOv8 image/video dataloader.
This class manages the loading and pre-processing of image and video data for YOLOv8. It supports loading from
various formats, including single image files, video files, and lists of image and video paths.
Attributes:
files (list): List of image and video file paths.
nf (int): Total number of files (images and videos).
video_flag (list): Flags indicating whether a file is a video (True) or an image (False).
mode (str): Current mode, 'image' or 'video'.
vid_stride (int): Stride for video frame-rate, defaults to 1.
bs (int): Batch size, set to 1 for this class.
cap (cv2.VideoCapture): Video capture object for OpenCV.
frame (int): Frame counter for video.
frames (int): Total number of frames in the video.
count (int): Counter for iteration, initialized at 0 during `__iter__()`.
Methods:
_new_video(path): Create a new cv2.VideoCapture object for a given video path.
"""
def __init__(self, path, batch=1, vid_stride=1):
"""Initialize the Dataloader and raise FileNotFoundError if file not found."""
parent = None
if isinstance(path, str) and Path(path).suffix == ".txt": # *.txt file with img/vid/dir on each line
parent = Path(path).parent
path = Path(path).read_text().splitlines() # list of sources
files = []
for p in sorted(path) if isinstance(path, (list, tuple)) else [path]:
a = str(Path(p).absolute()) # do not use .resolve() https://github.com/ultralytics/ultralytics/issues/2912
if "*" in a:
files.extend(sorted(glob.glob(a, recursive=True))) # glob
elif os.path.isdir(a):
files.extend(sorted(glob.glob(os.path.join(a, "*.*")))) # dir
elif os.path.isfile(a):
files.append(a) # files (absolute or relative to CWD)
elif parent and (parent / p).is_file():
files.append(str((parent / p).absolute())) # files (relative to *.txt file parent)
else:
raise FileNotFoundError(f"{p} does not exist")
images = [x for x in files if x.split(".")[-1].lower() in IMG_FORMATS]
videos = [x for x in files if x.split(".")[-1].lower() in VID_FORMATS]
ni, nv = len(images), len(videos)
self.files = images + videos
self.nf = ni + nv # number of files
self.ni = ni # number of images
self.video_flag = [False] * ni + [True] * nv
self.mode = "image"
self.vid_stride = vid_stride # video frame-rate stride
self.bs = batch
if any(videos):
self._new_video(videos[0]) # new video
else:
self.cap = None
if self.nf == 0:
raise FileNotFoundError(
f"No images or videos found in {p}. "
f"Supported formats are:\nimages: {IMG_FORMATS}\nvideos: {VID_FORMATS}"
)
def __iter__(self):
"""Returns an iterator object for VideoStream or ImageFolder."""
self.count = 0
return self
def __next__(self):
"""Returns the next batch of images or video frames along with their paths and metadata."""
paths, imgs, info = [], [], []
while len(imgs) < self.bs:
if self.count >= self.nf: # end of file list
if len(imgs) > 0:
return paths, imgs, info # return last partial batch
else:
raise StopIteration
path = self.files[self.count]
if self.video_flag[self.count]:
self.mode = "video"
if not self.cap or not self.cap.isOpened():
self._new_video(path)
for _ in range(self.vid_stride):
success = self.cap.grab()
if not success:
break # end of video or failure
if success:
success, im0 = self.cap.retrieve()
if success:
self.frame += 1
paths.append(path)
imgs.append(im0)
info.append(f"video {self.count + 1}/{self.nf} (frame {self.frame}/{self.frames}) {path}: ")
if self.frame == self.frames: # end of video
self.count += 1
self.cap.release()
else:
# Move to the next file if the current video ended or failed to open
self.count += 1
if self.cap:
self.cap.release()
if self.count < self.nf:
self._new_video(self.files[self.count])
else:
self.mode = "image"
im0 = cv2.imread(path) # BGR
if im0 is None:
raise FileNotFoundError(f"Image Not Found {path}")
paths.append(path)
imgs.append(im0)
info.append(f"image {self.count + 1}/{self.nf} {path}: ")
self.count += 1 # move to the next file
if self.count >= self.ni: # end of image list
break
return paths, imgs, info
def _new_video(self, path):
"""Creates a new video capture object for the given path."""
self.frame = 0
self.cap = cv2.VideoCapture(path)
self.fps = int(self.cap.get(cv2.CAP_PROP_FPS))
if not self.cap.isOpened():
raise FileNotFoundError(f"Failed to open video {path}")
self.frames = int(self.cap.get(cv2.CAP_PROP_FRAME_COUNT) / self.vid_stride)
def __len__(self):
"""Returns the number of batches in the object."""
return math.ceil(self.nf / self.bs) # number of files
class LoadPilAndNumpy:
"""
Load images from PIL and Numpy arrays for batch processing.
This class is designed to manage loading and pre-processing of image data from both PIL and Numpy formats.
It performs basic validation and format conversion to ensure that the images are in the required format for
downstream processing.
Attributes:
paths (list): List of image paths or autogenerated filenames.
im0 (list): List of images stored as Numpy arrays.
mode (str): Type of data being processed, defaults to 'image'.
bs (int): Batch size, equivalent to the length of `im0`.
Methods:
_single_check(im): Validate and format a single image to a Numpy array.
"""
def __init__(self, im0):
"""Initialize PIL and Numpy Dataloader."""
if not isinstance(im0, list):
im0 = [im0]
self.paths = [getattr(im, "filename", f"image{i}.jpg") for i, im in enumerate(im0)]
self.im0 = [self._single_check(im) for im in im0]
self.mode = "image"
self.bs = len(self.im0)
@staticmethod
def _single_check(im):
"""Validate and format an image to numpy array."""
assert isinstance(im, (Image.Image, np.ndarray)), f"Expected PIL/np.ndarray image type, but got {type(im)}"
if isinstance(im, Image.Image):
if im.mode != "RGB":
im = im.convert("RGB")
im = np.asarray(im)[:, :, ::-1]
im = np.ascontiguousarray(im) # contiguous
return im
def __len__(self):
"""Returns the length of the 'im0' attribute."""
return len(self.im0)
def __next__(self):
"""Returns batch paths, images, processed images, None, ''."""
if self.count == 1: # loop only once as it's batch inference
raise StopIteration
self.count += 1
return self.paths, self.im0, [""] * self.bs
def __iter__(self):
"""Enables iteration for class LoadPilAndNumpy."""
self.count = 0
return self
class LoadTensor:
"""
Load images from torch.Tensor data.
This class manages the loading and pre-processing of image data from PyTorch tensors for further processing.
Attributes:
im0 (torch.Tensor): The input tensor containing the image(s).
bs (int): Batch size, inferred from the shape of `im0`.
mode (str): Current mode, set to 'image'.
paths (list): List of image paths or filenames.
count (int): Counter for iteration, initialized at 0 during `__iter__()`.
Methods:
_single_check(im, stride): Validate and possibly modify the input tensor.
"""
def __init__(self, im0) -> None:
"""Initialize Tensor Dataloader."""
self.im0 = self._single_check(im0)
self.bs = self.im0.shape[0]
self.mode = "image"
self.paths = [getattr(im, "filename", f"image{i}.jpg") for i, im in enumerate(im0)]
@staticmethod
def _single_check(im, stride=32):
"""Validate and format an image to torch.Tensor."""
s = (
f"WARNING ⚠️ torch.Tensor inputs should be BCHW i.e. shape(1, 3, 640, 640) "
f"divisible by stride {stride}. Input shape{tuple(im.shape)} is incompatible."
)
if len(im.shape) != 4:
if len(im.shape) != 3:
raise ValueError(s)
LOGGER.warning(s)
im = im.unsqueeze(0)
if im.shape[2] % stride or im.shape[3] % stride:
raise ValueError(s)
if im.max() > 1.0 + torch.finfo(im.dtype).eps: # torch.float32 eps is 1.2e-07
LOGGER.warning(
f"WARNING ⚠️ torch.Tensor inputs should be normalized 0.0-1.0 but max value is {im.max()}. "
f"Dividing input by 255."
)
im = im.float() / 255.0
return im
def __iter__(self):
"""Returns an iterator object."""
self.count = 0
return self
def __next__(self):
"""Return next item in the iterator."""
if self.count == 1:
raise StopIteration
self.count += 1
return self.paths, self.im0, [""] * self.bs
def __len__(self):
"""Returns the batch size."""
return self.bs
def autocast_list(source):
"""Merges a list of source of different types into a list of numpy arrays or PIL images."""
files = []
for im in source:
if isinstance(im, (str, Path)): # filename or uri
files.append(Image.open(requests.get(im, stream=True).raw if str(im).startswith("http") else im))
elif isinstance(im, (Image.Image, np.ndarray)): # PIL or np Image
files.append(im)
else:
raise TypeError(
f"type {type(im).__name__} is not a supported Ultralytics prediction source type. \n"
f"See https://docs.ultralytics.com/modes/predict for supported source types."
)
return files
def get_best_youtube_url(url, use_pafy=True):
"""
Retrieves the URL of the best quality MP4 video stream from a given YouTube video.
This function uses the pafy or yt_dlp library to extract the video info from YouTube. It then finds the highest
quality MP4 format that has video codec but no audio codec, and returns the URL of this video stream.
Args:
url (str): The URL of the YouTube video.
use_pafy (bool): Use the pafy package, default=True, otherwise use yt_dlp package.
Returns:
(str): The URL of the best quality MP4 video stream, or None if no suitable stream is found.
"""
if use_pafy:
check_requirements(("pafy", "youtube_dl==2020.12.2"))
import pafy # noqa
return pafy.new(url).getbestvideo(preftype="mp4").url
else:
check_requirements("yt-dlp")
import yt_dlp
with yt_dlp.YoutubeDL({"quiet": True}) as ydl:
info_dict = ydl.extract_info(url, download=False) # extract info
for f in reversed(info_dict.get("formats", [])): # reversed because best is usually last
# Find a format with video codec, no audio, *.mp4 extension at least 1920x1080 size
good_size = (f.get("width") or 0) >= 1920 or (f.get("height") or 0) >= 1080
if good_size and f["vcodec"] != "none" and f["acodec"] == "none" and f["ext"] == "mp4":
return f.get("url")
# Define constants
LOADERS = (LoadStreams, LoadPilAndNumpy, LoadImagesAndVideos, LoadScreenshots)

18
ultralytics/data/scripts/download_weights.sh
View File

@ -1,18 +0,0 @@
#!/bin/bash
# Ultralytics YOLO 🚀, AGPL-3.0 license
# Download latest models from https://github.com/ultralytics/assets/releases
# Example usage: bash ultralytics/data/scripts/download_weights.sh
# parent
# └── weights
# ├── yolov8n.pt ← downloads here
# ├── yolov8s.pt
# └── ...
python - <<EOF
from ultralytics.utils.downloads import attempt_download_asset
assets = [f'yolov8{size}{suffix}.pt' for size in 'nsmlx' for suffix in ('', '-cls', '-seg', '-pose')]
for x in assets:
attempt_download_asset(f'weights/{x}')
EOF

60
ultralytics/data/scripts/get_coco.sh
View File

@ -1,60 +0,0 @@
#!/bin/bash
# Ultralytics YOLO 🚀, AGPL-3.0 license
# Download COCO 2017 dataset https://cocodataset.org
# Example usage: bash data/scripts/get_coco.sh
# parent
# ├── ultralytics
# └── datasets
# └── coco ← downloads here
# Arguments (optional) Usage: bash data/scripts/get_coco.sh --train --val --test --segments
if [ "$#" -gt 0 ]; then
for opt in "$@"; do
case "${opt}" in
--train) train=true ;;
--val) val=true ;;
--test) test=true ;;
--segments) segments=true ;;
--sama) sama=true ;;
esac
done
else
train=true
val=true
test=false
segments=false
sama=false
fi
# Download/unzip labels
d='../datasets' # unzip directory
url=https://github.com/ultralytics/yolov5/releases/download/v1.0/
if [ "$segments" == "true" ]; then
f='coco2017labels-segments.zip' # 169 MB
elif [ "$sama" == "true" ]; then
f='coco2017labels-segments-sama.zip' # 199 MB https://www.sama.com/sama-coco-dataset/
else
f='coco2017labels.zip' # 46 MB
fi
echo 'Downloading' $url$f ' ...'
curl -L $url$f -o $f -# && unzip -q $f -d $d && rm $f &
# Download/unzip images
d='../datasets/coco/images' # unzip directory
url=http://images.cocodataset.org/zips/
if [ "$train" == "true" ]; then
f='train2017.zip' # 19G, 118k images
echo 'Downloading' $url$f '...'
curl -L $url$f -o $f -# && unzip -q $f -d $d && rm $f &
fi
if [ "$val" == "true" ]; then
f='val2017.zip' # 1G, 5k images
echo 'Downloading' $url$f '...'
curl -L $url$f -o $f -# && unzip -q $f -d $d && rm $f &
fi
if [ "$test" == "true" ]; then
f='test2017.zip' # 7G, 41k images (optional)
echo 'Downloading' $url$f '...'
curl -L $url$f -o $f -# && unzip -q $f -d $d && rm $f &
fi
wait # finish background tasks

17
ultralytics/data/scripts/get_coco128.sh
View File

@ -1,17 +0,0 @@
#!/bin/bash
# Ultralytics YOLO 🚀, AGPL-3.0 license
# Download COCO128 dataset https://www.kaggle.com/ultralytics/coco128 (first 128 images from COCO train2017)
# Example usage: bash data/scripts/get_coco128.sh
# parent
# ├── ultralytics
# └── datasets
# └── coco128 ← downloads here
# Download/unzip images and labels
d='../datasets' # unzip directory
url=https://github.com/ultralytics/yolov5/releases/download/v1.0/
f='coco128.zip' # or 'coco128-segments.zip', 68 MB
echo 'Downloading' $url$f ' ...'
curl -L $url$f -o $f -# && unzip -q $f -d $d && rm $f &
wait # finish background tasks

51
ultralytics/data/scripts/get_imagenet.sh
View File

@ -1,51 +0,0 @@
#!/bin/bash
# Ultralytics YOLO 🚀, AGPL-3.0 license
# Download ILSVRC2012 ImageNet dataset https://image-net.org
# Example usage: bash data/scripts/get_imagenet.sh
# parent
# ├── ultralytics
# └── datasets
# └── imagenet ← downloads here
# Arguments (optional) Usage: bash data/scripts/get_imagenet.sh --train --val
if [ "$#" -gt 0 ]; then
for opt in "$@"; do
case "${opt}" in
--train) train=true ;;
--val) val=true ;;
esac
done
else
train=true
val=true
fi
# Make dir
d='../datasets/imagenet' # unzip directory
mkdir -p $d && cd $d
# Download/unzip train
if [ "$train" == "true" ]; then
wget https://image-net.org/data/ILSVRC/2012/ILSVRC2012_img_train.tar # download 138G, 1281167 images
mkdir train && mv ILSVRC2012_img_train.tar train/ && cd train
tar -xf ILSVRC2012_img_train.tar && rm -f ILSVRC2012_img_train.tar
find . -name "*.tar" | while read NAME; do
mkdir -p "${NAME%.tar}"
tar -xf "${NAME}" -C "${NAME%.tar}"
rm -f "${NAME}"
done
cd ..
fi
# Download/unzip val
if [ "$val" == "true" ]; then
wget https://image-net.org/data/ILSVRC/2012/ILSVRC2012_img_val.tar # download 6.3G, 50000 images
mkdir val && mv ILSVRC2012_img_val.tar val/ && cd val && tar -xf ILSVRC2012_img_val.tar
wget -qO- https://raw.githubusercontent.com/soumith/imagenetloader.torch/master/valprep.sh | bash # move into subdirs
fi
# Delete corrupted image (optional: PNG under JPEG name that may cause dataloaders to fail)
# rm train/n04266014/n04266014_10835.JPEG
# TFRecords (optional)
# wget https://raw.githubusercontent.com/tensorflow/models/master/research/slim/datasets/imagenet_lsvrc_2015_synsets.txt

288
ultralytics/data/split_dota.py
View File

@ -1,288 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import itertools
from glob import glob
from math import ceil
from pathlib import Path
import cv2
import numpy as np
from PIL import Image
from tqdm import tqdm
from ultralytics.data.utils import exif_size, img2label_paths
from ultralytics.utils.checks import check_requirements
check_requirements("shapely")
from shapely.geometry import Polygon
def bbox_iof(polygon1, bbox2, eps=1e-6):
"""
Calculate iofs between bbox1 and bbox2.
Args:
polygon1 (np.ndarray): Polygon coordinates, (n, 8).
bbox2 (np.ndarray): Bounding boxes, (n ,4).
"""
polygon1 = polygon1.reshape(-1, 4, 2)
lt_point = np.min(polygon1, axis=-2)
rb_point = np.max(polygon1, axis=-2)
bbox1 = np.concatenate([lt_point, rb_point], axis=-1)
lt = np.maximum(bbox1[:, None, :2], bbox2[..., :2])
rb = np.minimum(bbox1[:, None, 2:], bbox2[..., 2:])
wh = np.clip(rb - lt, 0, np.inf)
h_overlaps = wh[..., 0] * wh[..., 1]
l, t, r, b = (bbox2[..., i] for i in range(4))
polygon2 = np.stack([l, t, r, t, r, b, l, b], axis=-1).reshape(-1, 4, 2)
sg_polys1 = [Polygon(p) for p in polygon1]
sg_polys2 = [Polygon(p) for p in polygon2]
overlaps = np.zeros(h_overlaps.shape)
for p in zip(*np.nonzero(h_overlaps)):
overlaps[p] = sg_polys1[p[0]].intersection(sg_polys2[p[-1]]).area
unions = np.array([p.area for p in sg_polys1], dtype=np.float32)
unions = unions[..., None]
unions = np.clip(unions, eps, np.inf)
outputs = overlaps / unions
if outputs.ndim == 1:
outputs = outputs[..., None]
return outputs
def load_yolo_dota(data_root, split="train"):
"""
Load DOTA dataset.
Args:
data_root (str): Data root.
split (str): The split data set, could be train or val.
Notes:
The directory structure assumed for the DOTA dataset:
- data_root
- images
- train
- val
- labels
- train
- val
"""
assert split in ["train", "val"]
im_dir = Path(data_root) / "images" / split
assert im_dir.exists(), f"Can't find {im_dir}, please check your data root."
im_files = glob(str(Path(data_root) / "images" / split / "*"))
lb_files = img2label_paths(im_files)
annos = []
for im_file, lb_file in zip(im_files, lb_files):
w, h = exif_size(Image.open(im_file))
with open(lb_file) as f:
lb = [x.split() for x in f.read().strip().splitlines() if len(x)]
lb = np.array(lb, dtype=np.float32)
annos.append(dict(ori_size=(h, w), label=lb, filepath=im_file))
return annos
def get_windows(im_size, crop_sizes=[1024], gaps=[200], im_rate_thr=0.6, eps=0.01):
"""
Get the coordinates of windows.
Args:
im_size (tuple): Original image size, (h, w).
crop_sizes (List(int)): Crop size of windows.
gaps (List(int)): Gap between crops.
im_rate_thr (float): Threshold of windows areas divided by image ares.
"""
h, w = im_size
windows = []
for crop_size, gap in zip(crop_sizes, gaps):
assert crop_size > gap, f"invalid crop_size gap pair [{crop_size} {gap}]"
step = crop_size - gap
xn = 1 if w <= crop_size else ceil((w - crop_size) / step + 1)
xs = [step * i for i in range(xn)]
if len(xs) > 1 and xs[-1] + crop_size > w:
xs[-1] = w - crop_size
yn = 1 if h <= crop_size else ceil((h - crop_size) / step + 1)
ys = [step * i for i in range(yn)]
if len(ys) > 1 and ys[-1] + crop_size > h:
ys[-1] = h - crop_size
start = np.array(list(itertools.product(xs, ys)), dtype=np.int64)
stop = start + crop_size
windows.append(np.concatenate([start, stop], axis=1))
windows = np.concatenate(windows, axis=0)
im_in_wins = windows.copy()
im_in_wins[:, 0::2] = np.clip(im_in_wins[:, 0::2], 0, w)
im_in_wins[:, 1::2] = np.clip(im_in_wins[:, 1::2], 0, h)
im_areas = (im_in_wins[:, 2] - im_in_wins[:, 0]) * (im_in_wins[:, 3] - im_in_wins[:, 1])
win_areas = (windows[:, 2] - windows[:, 0]) * (windows[:, 3] - windows[:, 1])
im_rates = im_areas / win_areas
if not (im_rates > im_rate_thr).any():
max_rate = im_rates.max()
im_rates[abs(im_rates - max_rate) < eps] = 1
return windows[im_rates > im_rate_thr]
def get_window_obj(anno, windows, iof_thr=0.7):
"""Get objects for each window."""
h, w = anno["ori_size"]
label = anno["label"]
if len(label):
label[:, 1::2] *= w
label[:, 2::2] *= h
iofs = bbox_iof(label[:, 1:], windows)
# Unnormalized and misaligned coordinates
return [(label[iofs[:, i] >= iof_thr]) for i in range(len(windows))] # window_anns
else:
return [np.zeros((0, 9), dtype=np.float32) for _ in range(len(windows))] # window_anns
def crop_and_save(anno, windows, window_objs, im_dir, lb_dir):
"""
Crop images and save new labels.
Args:
anno (dict): Annotation dict, including `filepath`, `label`, `ori_size` as its keys.
windows (list): A list of windows coordinates.
window_objs (list): A list of labels inside each window.
im_dir (str): The output directory path of images.
lb_dir (str): The output directory path of labels.
Notes:
The directory structure assumed for the DOTA dataset:
- data_root
- images
- train
- val
- labels
- train
- val
"""
im = cv2.imread(anno["filepath"])
name = Path(anno["filepath"]).stem
for i, window in enumerate(windows):
x_start, y_start, x_stop, y_stop = window.tolist()
new_name = f"{name}__{x_stop - x_start}__{x_start}___{y_start}"
patch_im = im[y_start:y_stop, x_start:x_stop]
ph, pw = patch_im.shape[:2]
cv2.imwrite(str(Path(im_dir) / f"{new_name}.jpg"), patch_im)
label = window_objs[i]
if len(label) == 0:
continue
label[:, 1::2] -= x_start
label[:, 2::2] -= y_start
label[:, 1::2] /= pw
label[:, 2::2] /= ph
with open(Path(lb_dir) / f"{new_name}.txt", "w") as f:
for lb in label:
formatted_coords = ["{:.6g}".format(coord) for coord in lb[1:]]
f.write(f"{int(lb[0])} {' '.join(formatted_coords)}\n")
def split_images_and_labels(data_root, save_dir, split="train", crop_sizes=[1024], gaps=[200]):
"""
Split both images and labels.
Notes:
The directory structure assumed for the DOTA dataset:
- data_root
- images
- split
- labels
- split
and the output directory structure is:
- save_dir
- images
- split
- labels
- split
"""
im_dir = Path(save_dir) / "images" / split
im_dir.mkdir(parents=True, exist_ok=True)
lb_dir = Path(save_dir) / "labels" / split
lb_dir.mkdir(parents=True, exist_ok=True)
annos = load_yolo_dota(data_root, split=split)
for anno in tqdm(annos, total=len(annos), desc=split):
windows = get_windows(anno["ori_size"], crop_sizes, gaps)
window_objs = get_window_obj(anno, windows)
crop_and_save(anno, windows, window_objs, str(im_dir), str(lb_dir))
def split_trainval(data_root, save_dir, crop_size=1024, gap=200, rates=[1.0]):
"""
Split train and val set of DOTA.
Notes:
The directory structure assumed for the DOTA dataset:
- data_root
- images
- train
- val
- labels
- train
- val
and the output directory structure is:
- save_dir
- images
- train
- val
- labels
- train
- val
"""
crop_sizes, gaps = [], []
for r in rates:
crop_sizes.append(int(crop_size / r))
gaps.append(int(gap / r))
for split in ["train", "val"]:
split_images_and_labels(data_root, save_dir, split, crop_sizes, gaps)
def split_test(data_root, save_dir, crop_size=1024, gap=200, rates=[1.0]):
"""
Split test set of DOTA, labels are not included within this set.
Notes:
The directory structure assumed for the DOTA dataset:
- data_root
- images
- test
and the output directory structure is:
- save_dir
- images
- test
"""
crop_sizes, gaps = [], []
for r in rates:
crop_sizes.append(int(crop_size / r))
gaps.append(int(gap / r))
save_dir = Path(save_dir) / "images" / "test"
save_dir.mkdir(parents=True, exist_ok=True)
im_dir = Path(data_root) / "images" / "test"
assert im_dir.exists(), f"Can't find {im_dir}, please check your data root."
im_files = glob(str(im_dir / "*"))
for im_file in tqdm(im_files, total=len(im_files), desc="test"):
w, h = exif_size(Image.open(im_file))
windows = get_windows((h, w), crop_sizes=crop_sizes, gaps=gaps)
im = cv2.imread(im_file)
name = Path(im_file).stem
for window in windows:
x_start, y_start, x_stop, y_stop = window.tolist()
new_name = f"{name}__{x_stop - x_start}__{x_start}___{y_start}"
patch_im = im[y_start:y_stop, x_start:x_stop]
cv2.imwrite(str(save_dir / f"{new_name}.jpg"), patch_im)
if __name__ == "__main__":
split_trainval(data_root="DOTAv2", save_dir="DOTAv2-split")
split_test(data_root="DOTAv2", save_dir="DOTAv2-split")

651
ultralytics/data/utils.py
View File

@ -1,651 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import contextlib
import hashlib
import json
import os
import random
import subprocess
import time
import zipfile
from multiprocessing.pool import ThreadPool
from pathlib import Path
from tarfile import is_tarfile
import cv2
import numpy as np
from PIL import Image, ImageOps
from ultralytics.nn.autobackend import check_class_names
from ultralytics.utils import (
DATASETS_DIR,
LOGGER,
NUM_THREADS,
ROOT,
SETTINGS_YAML,
TQDM,
clean_url,
colorstr,
emojis,
yaml_load,
yaml_save,
)
from ultralytics.utils.checks import check_file, check_font, is_ascii
from ultralytics.utils.downloads import download, safe_download, unzip_file
from ultralytics.utils.ops import segments2boxes
HELP_URL = "See https://docs.ultralytics.com/datasets/detect for dataset formatting guidance."
IMG_FORMATS = {"bmp", "dng", "jpeg", "jpg", "mpo", "png", "tif", "tiff", "webp", "pfm"} # image suffixes
VID_FORMATS = {"asf", "avi", "gif", "m4v", "mkv", "mov", "mp4", "mpeg", "mpg", "ts", "wmv", "webm"} # video suffixes
PIN_MEMORY = str(os.getenv("PIN_MEMORY", True)).lower() == "true" # global pin_memory for dataloaders
def img2label_paths(img_paths):
"""Define label paths as a function of image paths."""
sa, sb = f"{os.sep}images{os.sep}", f"{os.sep}labels{os.sep}" # /images/, /labels/ substrings
return [sb.join(x.rsplit(sa, 1)).rsplit(".", 1)[0] + ".txt" for x in img_paths]
def get_hash(paths):
"""Returns a single hash value of a list of paths (files or dirs)."""
size = sum(os.path.getsize(p) for p in paths if os.path.exists(p)) # sizes
h = hashlib.sha256(str(size).encode()) # hash sizes
h.update("".join(paths).encode()) # hash paths
return h.hexdigest() # return hash
def exif_size(img: Image.Image):
"""Returns exif-corrected PIL size."""
s = img.size # (width, height)
if img.format == "JPEG": # only support JPEG images
with contextlib.suppress(Exception):
exif = img.getexif()
if exif:
rotation = exif.get(274, None) # the EXIF key for the orientation tag is 274
if rotation in [6, 8]: # rotation 270 or 90
s = s[1], s[0]
return s
def verify_image(args):
"""Verify one image."""
(im_file, cls), prefix = args
# Number (found, corrupt), message
nf, nc, msg = 0, 0, ""
try:
im = Image.open(im_file)
im.verify() # PIL verify
shape = exif_size(im) # image size
shape = (shape[1], shape[0]) # hw
assert (shape[0] > 9) & (shape[1] > 9), f"image size {shape} <10 pixels"
assert im.format.lower() in IMG_FORMATS, f"invalid image format {im.format}"
if im.format.lower() in ("jpg", "jpeg"):
with open(im_file, "rb") as f:
f.seek(-2, 2)
if f.read() != b"\xff\xd9": # corrupt JPEG
ImageOps.exif_transpose(Image.open(im_file)).save(im_file, "JPEG", subsampling=0, quality=100)
msg = f"{prefix}WARNING ⚠️ {im_file}: corrupt JPEG restored and saved"
nf = 1
except Exception as e:
nc = 1
msg = f"{prefix}WARNING ⚠️ {im_file}: ignoring corrupt image/label: {e}"
return (im_file, cls), nf, nc, msg
def verify_image_label(args):
"""Verify one image-label pair."""
im_file, lb_file, prefix, keypoint, num_cls, nkpt, ndim = args
# Number (missing, found, empty, corrupt), message, segments, keypoints
nm, nf, ne, nc, msg, segments, keypoints = 0, 0, 0, 0, "", [], None
try:
# Verify images
im = Image.open(im_file)
im.verify() # PIL verify
shape = exif_size(im) # image size
shape = (shape[1], shape[0]) # hw
assert (shape[0] > 9) & (shape[1] > 9), f"image size {shape} <10 pixels"
assert im.format.lower() in IMG_FORMATS, f"invalid image format {im.format}"
if im.format.lower() in ("jpg", "jpeg"):
with open(im_file, "rb") as f:
f.seek(-2, 2)
if f.read() != b"\xff\xd9": # corrupt JPEG
ImageOps.exif_transpose(Image.open(im_file)).save(im_file, "JPEG", subsampling=0, quality=100)
msg = f"{prefix}WARNING ⚠️ {im_file}: corrupt JPEG restored and saved"
# Verify labels
if os.path.isfile(lb_file):
nf = 1 # label found
with open(lb_file) as f:
lb = [x.split() for x in f.read().strip().splitlines() if len(x)]
if any(len(x) > 6 for x in lb) and (not keypoint): # is segment
classes = np.array([x[0] for x in lb], dtype=np.float32)
segments = [np.array(x[1:], dtype=np.float32).reshape(-1, 2) for x in lb] # (cls, xy1...)
lb = np.concatenate((classes.reshape(-1, 1), segments2boxes(segments)), 1) # (cls, xywh)
lb = np.array(lb, dtype=np.float32)
nl = len(lb)
if nl:
if keypoint:
assert lb.shape[1] == (5 + nkpt * ndim), f"labels require {(5 + nkpt * ndim)} columns each"
points = lb[:, 5:].reshape(-1, ndim)[:, :2]
else:
assert lb.shape[1] == 5, f"labels require 5 columns, {lb.shape[1]} columns detected"
points = lb[:, 1:]
assert points.max() <= 1, f"non-normalized or out of bounds coordinates {points[points > 1]}"
assert lb.min() >= 0, f"negative label values {lb[lb < 0]}"
# All labels
max_cls = lb[:, 0].max() # max label count
assert max_cls <= num_cls, (
f"Label class {int(max_cls)} exceeds dataset class count {num_cls}. "
f"Possible class labels are 0-{num_cls - 1}"
)
_, i = np.unique(lb, axis=0, return_index=True)
if len(i) < nl: # duplicate row check
lb = lb[i] # remove duplicates
if segments:
segments = [segments[x] for x in i]
msg = f"{prefix}WARNING ⚠️ {im_file}: {nl - len(i)} duplicate labels removed"
else:
ne = 1 # label empty
lb = np.zeros((0, (5 + nkpt * ndim) if keypoint else 5), dtype=np.float32)
else:
nm = 1 # label missing
lb = np.zeros((0, (5 + nkpt * ndim) if keypoints else 5), dtype=np.float32)
if keypoint:
keypoints = lb[:, 5:].reshape(-1, nkpt, ndim)
if ndim == 2:
kpt_mask = np.where((keypoints[..., 0] < 0) | (keypoints[..., 1] < 0), 0.0, 1.0).astype(np.float32)
keypoints = np.concatenate([keypoints, kpt_mask[..., None]], axis=-1) # (nl, nkpt, 3)
lb = lb[:, :5]
return im_file, lb, shape, segments, keypoints, nm, nf, ne, nc, msg
except Exception as e:
nc = 1
msg = f"{prefix}WARNING ⚠️ {im_file}: ignoring corrupt image/label: {e}"
return [None, None, None, None, None, nm, nf, ne, nc, msg]
def polygon2mask(imgsz, polygons, color=1, downsample_ratio=1):
"""
Convert a list of polygons to a binary mask of the specified image size.
Args:
imgsz (tuple): The size of the image as (height, width).
polygons (list[np.ndarray]): A list of polygons. Each polygon is an array with shape [N, M], where
N is the number of polygons, and M is the number of points such that M % 2 = 0.
color (int, optional): The color value to fill in the polygons on the mask. Defaults to 1.
downsample_ratio (int, optional): Factor by which to downsample the mask. Defaults to 1.
Returns:
(np.ndarray): A binary mask of the specified image size with the polygons filled in.
"""
mask = np.zeros(imgsz, dtype=np.uint8)
polygons = np.asarray(polygons, dtype=np.int32)
polygons = polygons.reshape((polygons.shape[0], -1, 2))
cv2.fillPoly(mask, polygons, color=color)
nh, nw = (imgsz[0] // downsample_ratio, imgsz[1] // downsample_ratio)
# Note: fillPoly first then resize is trying to keep the same loss calculation method when mask-ratio=1
return cv2.resize(mask, (nw, nh))
def polygons2masks(imgsz, polygons, color, downsample_ratio=1):
"""
Convert a list of polygons to a set of binary masks of the specified image size.
Args:
imgsz (tuple): The size of the image as (height, width).
polygons (list[np.ndarray]): A list of polygons. Each polygon is an array with shape [N, M], where
N is the number of polygons, and M is the number of points such that M % 2 = 0.
color (int): The color value to fill in the polygons on the masks.
downsample_ratio (int, optional): Factor by which to downsample each mask. Defaults to 1.
Returns:
(np.ndarray): A set of binary masks of the specified image size with the polygons filled in.
"""
return np.array([polygon2mask(imgsz, [x.reshape(-1)], color, downsample_ratio) for x in polygons])
def polygons2masks_overlap(imgsz, segments, downsample_ratio=1):
"""Return a (640, 640) overlap mask."""
masks = np.zeros(
(imgsz[0] // downsample_ratio, imgsz[1] // downsample_ratio),
dtype=np.int32 if len(segments) > 255 else np.uint8,
)
areas = []
ms = []
for si in range(len(segments)):
mask = polygon2mask(imgsz, [segments[si].reshape(-1)], downsample_ratio=downsample_ratio, color=1)
ms.append(mask)
areas.append(mask.sum())
areas = np.asarray(areas)
index = np.argsort(-areas)
ms = np.array(ms)[index]
for i in range(len(segments)):
mask = ms[i] * (i + 1)
masks = masks + mask
masks = np.clip(masks, a_min=0, a_max=i + 1)
return masks, index
def find_dataset_yaml(path: Path) -> Path:
"""
Find and return the YAML file associated with a Detect, Segment or Pose dataset.
This function searches for a YAML file at the root level of the provided directory first, and if not found, it
performs a recursive search. It prefers YAML files that have the same stem as the provided path. An AssertionError
is raised if no YAML file is found or if multiple YAML files are found.
Args:
path (Path): The directory path to search for the YAML file.
Returns:
(Path): The path of the found YAML file.
"""
files = list(path.glob("*.yaml")) or list(path.rglob("*.yaml")) # try root level first and then recursive
assert files, f"No YAML file found in '{path.resolve()}'"
if len(files) > 1:
files = [f for f in files if f.stem == path.stem] # prefer *.yaml files that match
assert len(files) == 1, f"Expected 1 YAML file in '{path.resolve()}', but found {len(files)}.\n{files}"
return files[0]
def check_det_dataset(dataset, autodownload=True):
"""
Download, verify, and/or unzip a dataset if not found locally.
This function checks the availability of a specified dataset, and if not found, it has the option to download and
unzip the dataset. It then reads and parses the accompanying YAML data, ensuring key requirements are met and also
resolves paths related to the dataset.
Args:
dataset (str): Path to the dataset or dataset descriptor (like a YAML file).
autodownload (bool, optional): Whether to automatically download the dataset if not found. Defaults to True.
Returns:
(dict): Parsed dataset information and paths.
"""
file = check_file(dataset)
# Download (optional)
extract_dir = ""
if zipfile.is_zipfile(file) or is_tarfile(file):
new_dir = safe_download(file, dir=DATASETS_DIR, unzip=True, delete=False)
file = find_dataset_yaml(DATASETS_DIR / new_dir)
extract_dir, autodownload = file.parent, False
# Read YAML
data = yaml_load(file, append_filename=True) # dictionary
# Checks
for k in "train", "val":
if k not in data:
if k != "val" or "validation" not in data:
raise SyntaxError(
emojis(f"{dataset} '{k}:' key missing ❌.\n'train' and 'val' are required in all data YAMLs.")
)
LOGGER.info("WARNING ⚠️ renaming data YAML 'validation' key to 'val' to match YOLO format.")
data["val"] = data.pop("validation") # replace 'validation' key with 'val' key
if "names" not in data and "nc" not in data:
raise SyntaxError(emojis(f"{dataset} key missing ❌.\n either 'names' or 'nc' are required in all data YAMLs."))
if "names" in data and "nc" in data and len(data["names"]) != data["nc"]:
raise SyntaxError(emojis(f"{dataset} 'names' length {len(data['names'])} and 'nc: {data['nc']}' must match."))
if "names" not in data:
data["names"] = [f"class_{i}" for i in range(data["nc"])]
else:
data["nc"] = len(data["names"])
data["names"] = check_class_names(data["names"])
# Resolve paths
path = Path(extract_dir or data.get("path") or Path(data.get("yaml_file", "")).parent) # dataset root
if not path.is_absolute():
path = (DATASETS_DIR / path).resolve()
# Set paths
data["path"] = path # download scripts
for k in "train", "val", "test":
if data.get(k): # prepend path
if isinstance(data[k], str):
x = (path / data[k]).resolve()
if not x.exists() and data[k].startswith("../"):
x = (path / data[k][3:]).resolve()
data[k] = str(x)
else:
data[k] = [str((path / x).resolve()) for x in data[k]]
# Parse YAML
val, s = (data.get(x) for x in ("val", "download"))
if val:
val = [Path(x).resolve() for x in (val if isinstance(val, list) else [val])] # val path
if not all(x.exists() for x in val):
name = clean_url(dataset) # dataset name with URL auth stripped
m = f"\nDataset '{name}' images not found ⚠️, missing path '{[x for x in val if not x.exists()][0]}'"
if s and autodownload:
LOGGER.warning(m)
else:
m += f"\nNote dataset download directory is '{DATASETS_DIR}'. You can update this in '{SETTINGS_YAML}'"
raise FileNotFoundError(m)
t = time.time()
r = None # success
if s.startswith("http") and s.endswith(".zip"): # URL
safe_download(url=s, dir=DATASETS_DIR, delete=True)
elif s.startswith("bash "): # bash script
LOGGER.info(f"Running {s} ...")
r = os.system(s)
else: # python script
exec(s, {"yaml": data})
dt = f"({round(time.time() - t, 1)}s)"
s = f"success ✅ {dt}, saved to {colorstr('bold', DATASETS_DIR)}" if r in (0, None) else f"failure {dt} ❌"
LOGGER.info(f"Dataset download {s}\n")
check_font("Arial.ttf" if is_ascii(data["names"]) else "Arial.Unicode.ttf") # download fonts
return data # dictionary
def check_cls_dataset(dataset, split=""):
"""
Checks a classification dataset such as Imagenet.
This function accepts a `dataset` name and attempts to retrieve the corresponding dataset information.
If the dataset is not found locally, it attempts to download the dataset from the internet and save it locally.
Args:
dataset (str | Path): The name of the dataset.
split (str, optional): The split of the dataset. Either 'val', 'test', or ''. Defaults to ''.
Returns:
(dict): A dictionary containing the following keys:
- 'train' (Path): The directory path containing the training set of the dataset.
- 'val' (Path): The directory path containing the validation set of the dataset.
- 'test' (Path): The directory path containing the test set of the dataset.
- 'nc' (int): The number of classes in the dataset.
- 'names' (dict): A dictionary of class names in the dataset.
"""
# Download (optional if dataset=https://file.zip is passed directly)
if str(dataset).startswith(("http:/", "https:/")):
dataset = safe_download(dataset, dir=DATASETS_DIR, unzip=True, delete=False)
elif Path(dataset).suffix in (".zip", ".tar", ".gz"):
file = check_file(dataset)
dataset = safe_download(file, dir=DATASETS_DIR, unzip=True, delete=False)
dataset = Path(dataset)
data_dir = (dataset if dataset.is_dir() else (DATASETS_DIR / dataset)).resolve()
if not data_dir.is_dir():
LOGGER.warning(f"\nDataset not found ⚠️, missing path {data_dir}, attempting download...")
t = time.time()
if str(dataset) == "imagenet":
subprocess.run(f"bash {ROOT / 'data/scripts/get_imagenet.sh'}", shell=True, check=True)
else:
url = f"https://github.com/ultralytics/yolov5/releases/download/v1.0/{dataset}.zip"
download(url, dir=data_dir.parent)
s = f"Dataset download success ✅ ({time.time() - t:.1f}s), saved to {colorstr('bold', data_dir)}\n"
LOGGER.info(s)
train_set = data_dir / "train"
val_set = (
data_dir / "val"
if (data_dir / "val").exists()
else data_dir / "validation"
if (data_dir / "validation").exists()
else None
) # data/test or data/val
test_set = data_dir / "test" if (data_dir / "test").exists() else None # data/val or data/test
if split == "val" and not val_set:
LOGGER.warning("WARNING ⚠️ Dataset 'split=val' not found, using 'split=test' instead.")
elif split == "test" and not test_set:
LOGGER.warning("WARNING ⚠️ Dataset 'split=test' not found, using 'split=val' instead.")
nc = len([x for x in (data_dir / "train").glob("*") if x.is_dir()]) # number of classes
names = [x.name for x in (data_dir / "train").iterdir() if x.is_dir()] # class names list
names = dict(enumerate(sorted(names)))
# Print to console
for k, v in {"train": train_set, "val": val_set, "test": test_set}.items():
prefix = f'{colorstr(f"{k}:")} {v}...'
if v is None:
LOGGER.info(prefix)
else:
files = [path for path in v.rglob("*.*") if path.suffix[1:].lower() in IMG_FORMATS]
nf = len(files) # number of files
nd = len({file.parent for file in files}) # number of directories
if nf == 0:
if k == "train":
raise FileNotFoundError(emojis(f"{dataset} '{k}:' no training images found ❌ "))
else:
LOGGER.warning(f"{prefix} found {nf} images in {nd} classes: WARNING ⚠️ no images found")
elif nd != nc:
LOGGER.warning(f"{prefix} found {nf} images in {nd} classes: ERROR ❌️ requires {nc} classes, not {nd}")
else:
LOGGER.info(f"{prefix} found {nf} images in {nd} classes ✅ ")
return {"train": train_set, "val": val_set, "test": test_set, "nc": nc, "names": names}
class HUBDatasetStats:
"""
A class for generating HUB dataset JSON and `-hub` dataset directory.
Args:
path (str): Path to data.yaml or data.zip (with data.yaml inside data.zip). Default is 'coco8.yaml'.
task (str): Dataset task. Options are 'detect', 'segment', 'pose', 'classify'. Default is 'detect'.
autodownload (bool): Attempt to download dataset if not found locally. Default is False.
Example:
Download *.zip files from https://github.com/ultralytics/hub/tree/main/example_datasets
i.e. https://github.com/ultralytics/hub/raw/main/example_datasets/coco8.zip for coco8.zip.
```python
from ultralytics.data.utils import HUBDatasetStats
stats = HUBDatasetStats('path/to/coco8.zip', task='detect') # detect dataset
stats = HUBDatasetStats('path/to/coco8-seg.zip', task='segment') # segment dataset
stats = HUBDatasetStats('path/to/coco8-pose.zip', task='pose') # pose dataset
stats = HUBDatasetStats('path/to/imagenet10.zip', task='classify') # classification dataset
stats.get_json(save=True)
stats.process_images()
```
"""
def __init__(self, path="coco8.yaml", task="detect", autodownload=False):
"""Initialize class."""
path = Path(path).resolve()
LOGGER.info(f"Starting HUB dataset checks for {path}....")
self.task = task # detect, segment, pose, classify
if self.task == "classify":
unzip_dir = unzip_file(path)
data = check_cls_dataset(unzip_dir)
data["path"] = unzip_dir
else: # detect, segment, pose
_, data_dir, yaml_path = self._unzip(Path(path))
try:
# Load YAML with checks
data = yaml_load(yaml_path)
data["path"] = "" # strip path since YAML should be in dataset root for all HUB datasets
yaml_save(yaml_path, data)
data = check_det_dataset(yaml_path, autodownload) # dict
data["path"] = data_dir # YAML path should be set to '' (relative) or parent (absolute)
except Exception as e:
raise Exception("error/HUB/dataset_stats/init") from e
self.hub_dir = Path(f'{data["path"]}-hub')
self.im_dir = self.hub_dir / "images"
self.stats = {"nc": len(data["names"]), "names": list(data["names"].values())} # statistics dictionary
self.data = data
@staticmethod
def _unzip(path):
"""Unzip data.zip."""
if not str(path).endswith(".zip"): # path is data.yaml
return False, None, path
unzip_dir = unzip_file(path, path=path.parent)
assert unzip_dir.is_dir(), (
f"Error unzipping {path}, {unzip_dir} not found. " f"path/to/abc.zip MUST unzip to path/to/abc/"
)
return True, str(unzip_dir), find_dataset_yaml(unzip_dir) # zipped, data_dir, yaml_path
def _hub_ops(self, f):
"""Saves a compressed image for HUB previews."""
compress_one_image(f, self.im_dir / Path(f).name) # save to dataset-hub
def get_json(self, save=False, verbose=False):
"""Return dataset JSON for Ultralytics HUB."""
def _round(labels):
"""Update labels to integer class and 4 decimal place floats."""
if self.task == "detect":
coordinates = labels["bboxes"]
elif self.task == "segment":
coordinates = [x.flatten() for x in labels["segments"]]
elif self.task == "pose":
n = labels["keypoints"].shape[0]
coordinates = np.concatenate((labels["bboxes"], labels["keypoints"].reshape(n, -1)), 1)
else:
raise ValueError("Undefined dataset task.")
zipped = zip(labels["cls"], coordinates)
return [[int(c[0]), *(round(float(x), 4) for x in points)] for c, points in zipped]
for split in "train", "val", "test":
self.stats[split] = None # predefine
path = self.data.get(split)
# Check split
if path is None: # no split
continue
files = [f for f in Path(path).rglob("*.*") if f.suffix[1:].lower() in IMG_FORMATS] # image files in split
if not files: # no images
continue
# Get dataset statistics
if self.task == "classify":
from torchvision.datasets import ImageFolder
dataset = ImageFolder(self.data[split])
x = np.zeros(len(dataset.classes)).astype(int)
for im in dataset.imgs:
x[im[1]] += 1
self.stats[split] = {
"instance_stats": {"total": len(dataset), "per_class": x.tolist()},
"image_stats": {"total": len(dataset), "unlabelled": 0, "per_class": x.tolist()},
"labels": [{Path(k).name: v} for k, v in dataset.imgs],
}
else:
from ultralytics.data import YOLODataset
dataset = YOLODataset(img_path=self.data[split], data=self.data, task=self.task)
x = np.array(
[
np.bincount(label["cls"].astype(int).flatten(), minlength=self.data["nc"])
for label in TQDM(dataset.labels, total=len(dataset), desc="Statistics")
]
) # shape(128x80)
self.stats[split] = {
"instance_stats": {"total": int(x.sum()), "per_class": x.sum(0).tolist()},
"image_stats": {
"total": len(dataset),
"unlabelled": int(np.all(x == 0, 1).sum()),
"per_class": (x > 0).sum(0).tolist(),
},
"labels": [{Path(k).name: _round(v)} for k, v in zip(dataset.im_files, dataset.labels)],
}
# Save, print and return
if save:
self.hub_dir.mkdir(parents=True, exist_ok=True) # makes dataset-hub/
stats_path = self.hub_dir / "stats.json"
LOGGER.info(f"Saving {stats_path.resolve()}...")
with open(stats_path, "w") as f:
json.dump(self.stats, f) # save stats.json
if verbose:
LOGGER.info(json.dumps(self.stats, indent=2, sort_keys=False))
return self.stats
def process_images(self):
"""Compress images for Ultralytics HUB."""
from ultralytics.data import YOLODataset # ClassificationDataset
self.im_dir.mkdir(parents=True, exist_ok=True) # makes dataset-hub/images/
for split in "train", "val", "test":
if self.data.get(split) is None:
continue
dataset = YOLODataset(img_path=self.data[split], data=self.data)
with ThreadPool(NUM_THREADS) as pool:
for _ in TQDM(pool.imap(self._hub_ops, dataset.im_files), total=len(dataset), desc=f"{split} images"):
pass
LOGGER.info(f"Done. All images saved to {self.im_dir}")
return self.im_dir
def compress_one_image(f, f_new=None, max_dim=1920, quality=50):
"""
Compresses a single image file to reduced size while preserving its aspect ratio and quality using either the Python
Imaging Library (PIL) or OpenCV library. If the input image is smaller than the maximum dimension, it will not be
resized.
Args:
f (str): The path to the input image file.
f_new (str, optional): The path to the output image file. If not specified, the input file will be overwritten.
max_dim (int, optional): The maximum dimension (width or height) of the output image. Default is 1920 pixels.
quality (int, optional): The image compression quality as a percentage. Default is 50%.
Example:
```python
from pathlib import Path
from ultralytics.data.utils import compress_one_image
for f in Path('path/to/dataset').rglob('*.jpg'):
compress_one_image(f)
```
"""
try: # use PIL
im = Image.open(f)
r = max_dim / max(im.height, im.width) # ratio
if r < 1.0: # image too large
im = im.resize((int(im.width * r), int(im.height * r)))
im.save(f_new or f, "JPEG", quality=quality, optimize=True) # save
except Exception as e: # use OpenCV
LOGGER.info(f"WARNING ⚠️ HUB ops PIL failure {f}: {e}")
im = cv2.imread(f)
im_height, im_width = im.shape[:2]
r = max_dim / max(im_height, im_width) # ratio
if r < 1.0: # image too large
im = cv2.resize(im, (int(im_width * r), int(im_height * r)), interpolation=cv2.INTER_AREA)
cv2.imwrite(str(f_new or f), im)
def autosplit(path=DATASETS_DIR / "coco8/images", weights=(0.9, 0.1, 0.0), annotated_only=False):
"""
Automatically split a dataset into train/val/test splits and save the resulting splits into autosplit_*.txt files.
Args:
path (Path, optional): Path to images directory. Defaults to DATASETS_DIR / 'coco8/images'.
weights (list | tuple, optional): Train, validation, and test split fractions. Defaults to (0.9, 0.1, 0.0).
annotated_only (bool, optional): If True, only images with an associated txt file are used. Defaults to False.
Example:
```python
from ultralytics.data.utils import autosplit
autosplit()
```
"""
path = Path(path) # images dir
files = sorted(x for x in path.rglob("*.*") if x.suffix[1:].lower() in IMG_FORMATS) # image files only
n = len(files) # number of files
random.seed(0) # for reproducibility
indices = random.choices([0, 1, 2], weights=weights, k=n) # assign each image to a split
txt = ["autosplit_train.txt", "autosplit_val.txt", "autosplit_test.txt"] # 3 txt files
for x in txt:
if (path.parent / x).exists():
(path.parent / x).unlink() # remove existing
LOGGER.info(f"Autosplitting images from {path}" + ", using *.txt labeled images only" * annotated_only)
for i, img in TQDM(zip(indices, files), total=n):
if not annotated_only or Path(img2label_paths([str(img)])[0]).exists(): # check label
with open(path.parent / txt[i], "a") as f:
f.write(f"./{img.relative_to(path.parent).as_posix()}" + "\n") # add image to txt file

1
ultralytics/engine/__init__.py
View File

@ -1 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license

1145
ultralytics/engine/exporter.py
View File

File diff suppressed because it is too large Load Diff

823
ultralytics/engine/model.py
View File

@ -1,823 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import inspect
import sys
from pathlib import Path
from typing import Union
import numpy as np
import torch
from ultralytics.cfg import TASK2DATA, get_cfg, get_save_dir
from ultralytics.hub.utils import HUB_WEB_ROOT
from ultralytics.nn.tasks import attempt_load_one_weight, guess_model_task, nn, yaml_model_load
from ultralytics.utils import ASSETS, DEFAULT_CFG_DICT, LOGGER, RANK, SETTINGS, callbacks, checks, emojis, yaml_load
class Model(nn.Module):
"""
A base class for implementing YOLO models, unifying APIs across different model types.
This class provides a common interface for various operations related to YOLO models, such as training,
validation, prediction, exporting, and benchmarking. It handles different types of models, including those
loaded from local files, Ultralytics HUB, or Triton Server. The class is designed to be flexible and
extendable for different tasks and model configurations.
Args:
model (Union[str, Path], optional): Path or name of the model to load or create. This can be a local file
path, a model name from Ultralytics HUB, or a Triton Server model. Defaults to 'yolov8n.pt'.
task (Any, optional): The task type associated with the YOLO model. This can be used to specify the model's
application domain, such as object detection, segmentation, etc. Defaults to None.
verbose (bool, optional): If True, enables verbose output during the model's operations. Defaults to False.
Attributes:
callbacks (dict): A dictionary of callback functions for various events during model operations.
predictor (BasePredictor): The predictor object used for making predictions.
model (nn.Module): The underlying PyTorch model.
trainer (BaseTrainer): The trainer object used for training the model.
ckpt (dict): The checkpoint data if the model is loaded from a *.pt file.
cfg (str): The configuration of the model if loaded from a *.yaml file.
ckpt_path (str): The path to the checkpoint file.
overrides (dict): A dictionary of overrides for model configuration.
metrics (dict): The latest training/validation metrics.
session (HUBTrainingSession): The Ultralytics HUB session, if applicable.
task (str): The type of task the model is intended for.
model_name (str): The name of the model.
Methods:
__call__: Alias for the predict method, enabling the model instance to be callable.
_new: Initializes a new model based on a configuration file.
_load: Loads a model from a checkpoint file.
_check_is_pytorch_model: Ensures that the model is a PyTorch model.
reset_weights: Resets the model's weights to their initial state.
load: Loads model weights from a specified file.
save: Saves the current state of the model to a file.
info: Logs or returns information about the model.
fuse: Fuses Conv2d and BatchNorm2d layers for optimized inference.
predict: Performs object detection predictions.
track: Performs object tracking.
val: Validates the model on a dataset.
benchmark: Benchmarks the model on various export formats.
export: Exports the model to different formats.
train: Trains the model on a dataset.
tune: Performs hyperparameter tuning.
_apply: Applies a function to the model's tensors.
add_callback: Adds a callback function for an event.
clear_callback: Clears all callbacks for an event.
reset_callbacks: Resets all callbacks to their default functions.
_get_hub_session: Retrieves or creates an Ultralytics HUB session.
is_triton_model: Checks if a model is a Triton Server model.
is_hub_model: Checks if a model is an Ultralytics HUB model.
_reset_ckpt_args: Resets checkpoint arguments when loading a PyTorch model.
_smart_load: Loads the appropriate module based on the model task.
task_map: Provides a mapping from model tasks to corresponding classes.
Raises:
FileNotFoundError: If the specified model file does not exist or is inaccessible.
ValueError: If the model file or configuration is invalid or unsupported.
ImportError: If required dependencies for specific model types (like HUB SDK) are not installed.
TypeError: If the model is not a PyTorch model when required.
AttributeError: If required attributes or methods are not implemented or available.
NotImplementedError: If a specific model task or mode is not supported.
"""
def __init__(
self,
model: Union[str, Path] = "yolov8n.pt",
task: str = None,
verbose: bool = False,
) -> None:
"""
Initializes a new instance of the YOLO model class.
This constructor sets up the model based on the provided model path or name. It handles various types of model
sources, including local files, Ultralytics HUB models, and Triton Server models. The method initializes several
important attributes of the model and prepares it for operations like training, prediction, or export.
Args:
model (Union[str, Path], optional): The path or model file to load or create. This can be a local
file path, a model name from Ultralytics HUB, or a Triton Server model. Defaults to 'yolov8n.pt'.
task (Any, optional): The task type associated with the YOLO model, specifying its application domain.
Defaults to None.
verbose (bool, optional): If True, enables verbose output during the model's initialization and subsequent
operations. Defaults to False.
Raises:
FileNotFoundError: If the specified model file does not exist or is inaccessible.
ValueError: If the model file or configuration is invalid or unsupported.
ImportError: If required dependencies for specific model types (like HUB SDK) are not installed.
"""
super().__init__()
self.callbacks = callbacks.get_default_callbacks()
self.predictor = None # reuse predictor
self.model = None # model object
self.trainer = None # trainer object
self.ckpt = None # if loaded from *.pt
self.cfg = None # if loaded from *.yaml
self.ckpt_path = None
self.overrides = {} # overrides for trainer object
self.metrics = None # validation/training metrics
self.session = None # HUB session
self.task = task # task type
model = str(model).strip()
# Check if Ultralytics HUB model from https://hub.ultralytics.com
if self.is_hub_model(model):
# Fetch model from HUB
checks.check_requirements("hub-sdk>=0.0.6")
self.session = self._get_hub_session(model)
model = self.session.model_file
# Check if Triton Server model
elif self.is_triton_model(model):
self.model_name = self.model = model
self.task = task
return
# Load or create new YOLO model
if Path(model).suffix in (".yaml", ".yml"):
self._new(model, task=task, verbose=verbose)
else:
self._load(model, task=task)
def __call__(
self,
source: Union[str, Path, int, list, tuple, np.ndarray, torch.Tensor] = None,
stream: bool = False,
**kwargs,
) -> list:
"""
An alias for the predict method, enabling the model instance to be callable.
This method simplifies the process of making predictions by allowing the model instance to be called directly
with the required arguments for prediction.
Args:
source (str | Path | int | PIL.Image | np.ndarray, optional): The source of the image for making
predictions. Accepts various types, including file paths, URLs, PIL images, and numpy arrays.
Defaults to None.
stream (bool, optional): If True, treats the input source as a continuous stream for predictions.
Defaults to False.
**kwargs (any): Additional keyword arguments for configuring the prediction process.
Returns:
(List[ultralytics.engine.results.Results]): A list of prediction results, encapsulated in the Results class.
"""
return self.predict(source, stream, **kwargs)
@staticmethod
def _get_hub_session(model: str):
"""Creates a session for Hub Training."""
from ultralytics.hub.session import HUBTrainingSession
session = HUBTrainingSession(model)
return session if session.client.authenticated else None
@staticmethod
def is_triton_model(model: str) -> bool:
"""Is model a Triton Server URL string, i.e. <scheme>://<netloc>/<endpoint>/<task_name>"""
from urllib.parse import urlsplit
url = urlsplit(model)
return url.netloc and url.path and url.scheme in {"http", "grpc"}
@staticmethod
def is_hub_model(model: str) -> bool:
"""Check if the provided model is a HUB model."""
return any(
(
model.startswith(f"{HUB_WEB_ROOT}/models/"), # i.e. https://hub.ultralytics.com/models/MODEL_ID
[len(x) for x in model.split("_")] == [42, 20], # APIKEY_MODEL
len(model) == 20 and not Path(model).exists() and all(x not in model for x in "./\\"), # MODEL
)
)
def _new(self, cfg: str, task=None, model=None, verbose=False) -> None:
"""
Initializes a new model and infers the task type from the model definitions.
Args:
cfg (str): model configuration file
task (str | None): model task
model (BaseModel): Customized model.
verbose (bool): display model info on load
"""
cfg_dict = yaml_model_load(cfg)
self.cfg = cfg
self.task = task or guess_model_task(cfg_dict)
self.model = (model or self._smart_load("model"))(cfg_dict, verbose=verbose and RANK == -1) # build model
self.overrides["model"] = self.cfg
self.overrides["task"] = self.task
# Below added to allow export from YAMLs
self.model.args = {**DEFAULT_CFG_DICT, **self.overrides} # combine default and model args (prefer model args)
self.model.task = self.task
self.model_name = cfg
def _load(self, weights: str, task=None) -> None:
"""
Initializes a new model and infers the task type from the model head.
Args:
weights (str): model checkpoint to be loaded
task (str | None): model task
"""
if weights.lower().startswith(("https://", "http://", "rtsp://", "rtmp://", "tcp://")):
weights = checks.check_file(weights) # automatically download and return local filename
weights = checks.check_model_file_from_stem(weights) # add suffix, i.e. yolov8n -> yolov8n.pt
if Path(weights).suffix == ".pt":
self.model, self.ckpt = attempt_load_one_weight(weights)
self.task = self.model.args["task"]
self.overrides = self.model.args = self._reset_ckpt_args(self.model.args)
self.ckpt_path = self.model.pt_path
else:
weights = checks.check_file(weights) # runs in all cases, not redundant with above call
self.model, self.ckpt = weights, None
self.task = task or guess_model_task(weights)
self.ckpt_path = weights
self.overrides["model"] = weights
self.overrides["task"] = self.task
self.model_name = weights
def _check_is_pytorch_model(self) -> None:
"""Raises TypeError is model is not a PyTorch model."""
pt_str = isinstance(self.model, (str, Path)) and Path(self.model).suffix == ".pt"
pt_module = isinstance(self.model, nn.Module)
if not (pt_module or pt_str):
raise TypeError(
f"model='{self.model}' should be a *.pt PyTorch model to run this method, but is a different format. "
f"PyTorch models can train, val, predict and export, i.e. 'model.train(data=...)', but exported "
f"formats like ONNX, TensorRT etc. only support 'predict' and 'val' modes, "
f"i.e. 'yolo predict model=yolov8n.onnx'.\nTo run CUDA or MPS inference please pass the device "
f"argument directly in your inference command, i.e. 'model.predict(source=..., device=0)'"
)
def reset_weights(self) -> "Model":
"""
Resets the model parameters to randomly initialized values, effectively discarding all training information.
This method iterates through all modules in the model and resets their parameters if they have a
'reset_parameters' method. It also ensures that all parameters have 'requires_grad' set to True, enabling them
to be updated during training.
Returns:
self (ultralytics.engine.model.Model): The instance of the class with reset weights.
Raises:
AssertionError: If the model is not a PyTorch model.
"""
self._check_is_pytorch_model()
for m in self.model.modules():
if hasattr(m, "reset_parameters"):
m.reset_parameters()
for p in self.model.parameters():
p.requires_grad = True
return self
def load(self, weights: Union[str, Path] = "yolov8n.pt") -> "Model":
"""
Loads parameters from the specified weights file into the model.
This method supports loading weights from a file or directly from a weights object. It matches parameters by
name and shape and transfers them to the model.
Args:
weights (str | Path): Path to the weights file or a weights object. Defaults to 'yolov8n.pt'.
Returns:
self (ultralytics.engine.model.Model): The instance of the class with loaded weights.
Raises:
AssertionError: If the model is not a PyTorch model.
"""
self._check_is_pytorch_model()
if isinstance(weights, (str, Path)):
weights, self.ckpt = attempt_load_one_weight(weights)
self.model.load(weights)
return self
def save(self, filename: Union[str, Path] = "saved_model.pt", use_dill=True) -> None:
"""
Saves the current model state to a file.
This method exports the model's checkpoint (ckpt) to the specified filename.
Args:
filename (str | Path): The name of the file to save the model to. Defaults to 'saved_model.pt'.
use_dill (bool): Whether to try using dill for serialization if available. Defaults to True.
Raises:
AssertionError: If the model is not a PyTorch model.
"""
self._check_is_pytorch_model()
from ultralytics import __version__
from datetime import datetime
updates = {
"date": datetime.now().isoformat(),
"version": __version__,
"license": "AGPL-3.0 License (https://ultralytics.com/license)",
"docs": "https://docs.ultralytics.com",
}
torch.save({**self.ckpt, **updates}, filename, use_dill=use_dill)
def info(self, detailed: bool = False, verbose: bool = True):
"""
Logs or returns model information.
This method provides an overview or detailed information about the model, depending on the arguments passed.
It can control the verbosity of the output.
Args:
detailed (bool): If True, shows detailed information about the model. Defaults to False.
verbose (bool): If True, prints the information. If False, returns the information. Defaults to True.
Returns:
(list): Various types of information about the model, depending on the 'detailed' and 'verbose' parameters.
Raises:
AssertionError: If the model is not a PyTorch model.
"""
self._check_is_pytorch_model()
return self.model.info(detailed=detailed, verbose=verbose)
def fuse(self):
"""
Fuses Conv2d and BatchNorm2d layers in the model.
This method optimizes the model by fusing Conv2d and BatchNorm2d layers, which can improve inference speed.
Raises:
AssertionError: If the model is not a PyTorch model.
"""
self._check_is_pytorch_model()
self.model.fuse()
def embed(
self,
source: Union[str, Path, int, list, tuple, np.ndarray, torch.Tensor] = None,
stream: bool = False,
**kwargs,
) -> list:
"""
Generates image embeddings based on the provided source.
This method is a wrapper around the 'predict()' method, focusing on generating embeddings from an image source.
It allows customization of the embedding process through various keyword arguments.
Args:
source (str | int | PIL.Image | np.ndarray): The source of the image for generating embeddings.
The source can be a file path, URL, PIL image, numpy array, etc. Defaults to None.
stream (bool): If True, predictions are streamed. Defaults to False.
**kwargs (any): Additional keyword arguments for configuring the embedding process.
Returns:
(List[torch.Tensor]): A list containing the image embeddings.
Raises:
AssertionError: If the model is not a PyTorch model.
"""
if not kwargs.get("embed"):
kwargs["embed"] = [len(self.model.model) - 2] # embed second-to-last layer if no indices passed
return self.predict(source, stream, **kwargs)
def predict(
self,
source: Union[str, Path, int, list, tuple, np.ndarray, torch.Tensor] = None,
stream: bool = False,
predictor=None,
**kwargs,
) -> list:
"""
Performs predictions on the given image source using the YOLO model.
This method facilitates the prediction process, allowing various configurations through keyword arguments.
It supports predictions with custom predictors or the default predictor method. The method handles different
types of image sources and can operate in a streaming mode. It also provides support for SAM-type models
through 'prompts'.
The method sets up a new predictor if not already present and updates its arguments with each call.
It also issues a warning and uses default assets if the 'source' is not provided. The method determines if it
is being called from the command line interface and adjusts its behavior accordingly, including setting defaults
for confidence threshold and saving behavior.
Args:
source (str | int | PIL.Image | np.ndarray, optional): The source of the image for making predictions.
Accepts various types, including file paths, URLs, PIL images, and numpy arrays. Defaults to ASSETS.
stream (bool, optional): Treats the input source as a continuous stream for predictions. Defaults to False.
predictor (BasePredictor, optional): An instance of a custom predictor class for making predictions.
If None, the method uses a default predictor. Defaults to None.
**kwargs (any): Additional keyword arguments for configuring the prediction process. These arguments allow
for further customization of the prediction behavior.
Returns:
(List[ultralytics.engine.results.Results]): A list of prediction results, encapsulated in the Results class.
Raises:
AttributeError: If the predictor is not properly set up.
"""
if source is None:
source = ASSETS
LOGGER.warning(f"WARNING ⚠️ 'source' is missing. Using 'source={source}'.")
is_cli = (sys.argv[0].endswith("yolo") or sys.argv[0].endswith("ultralytics")) and any(
x in sys.argv for x in ("predict", "track", "mode=predict", "mode=track")
)
custom = {"conf": 0.25, "batch": 1, "save": is_cli, "mode": "predict"} # method defaults
args = {**self.overrides, **custom, **kwargs} # highest priority args on the right
prompts = args.pop("prompts", None) # for SAM-type models
if not self.predictor:
self.predictor = predictor or self._smart_load("predictor")(overrides=args, _callbacks=self.callbacks)
self.predictor.setup_model(model=self.model, verbose=is_cli)
else: # only update args if predictor is already setup
self.predictor.args = get_cfg(self.predictor.args, args)
if "project" in args or "name" in args:
self.predictor.save_dir = get_save_dir(self.predictor.args)
if prompts and hasattr(self.predictor, "set_prompts"): # for SAM-type models
self.predictor.set_prompts(prompts)
return self.predictor.predict_cli(source=source) if is_cli else self.predictor(source=source, stream=stream)
def track(
self,
source: Union[str, Path, int, list, tuple, np.ndarray, torch.Tensor] = None,
stream: bool = False,
persist: bool = False,
**kwargs,
) -> list:
"""
Conducts object tracking on the specified input source using the registered trackers.
This method performs object tracking using the model's predictors and optionally registered trackers. It is
capable of handling different types of input sources such as file paths or video streams. The method supports
customization of the tracking process through various keyword arguments. It registers trackers if they are not
already present and optionally persists them based on the 'persist' flag.
The method sets a default confidence threshold specifically for ByteTrack-based tracking, which requires low
confidence predictions as input. The tracking mode is explicitly set in the keyword arguments.
Args:
source (str, optional): The input source for object tracking. It can be a file path, URL, or video stream.
stream (bool, optional): Treats the input source as a continuous video stream. Defaults to False.
persist (bool, optional): Persists the trackers between different calls to this method. Defaults to False.
**kwargs (any): Additional keyword arguments for configuring the tracking process. These arguments allow
for further customization of the tracking behavior.
Returns:
(List[ultralytics.engine.results.Results]): A list of tracking results, encapsulated in the Results class.
Raises:
AttributeError: If the predictor does not have registered trackers.
"""
if not hasattr(self.predictor, "trackers"):
from ultralytics.trackers import register_tracker
register_tracker(self, persist)
kwargs["conf"] = kwargs.get("conf") or 0.1 # ByteTrack-based method needs low confidence predictions as input
kwargs["batch"] = kwargs.get("batch") or 1 # batch-size 1 for tracking in videos
kwargs["mode"] = "track"
return self.predict(source=source, stream=stream, **kwargs)
def val(
self,
validator=None,
**kwargs,
):
"""
Validates the model using a specified dataset and validation configuration.
This method facilitates the model validation process, allowing for a range of customization through various
settings and configurations. It supports validation with a custom validator or the default validation approach.
The method combines default configurations, method-specific defaults, and user-provided arguments to configure
the validation process. After validation, it updates the model's metrics with the results obtained from the
validator.
The method supports various arguments that allow customization of the validation process. For a comprehensive
list of all configurable options, users should refer to the 'configuration' section in the documentation.
Args:
validator (BaseValidator, optional): An instance of a custom validator class for validating the model. If
None, the method uses a default validator. Defaults to None.
**kwargs (any): Arbitrary keyword arguments representing the validation configuration. These arguments are
used to customize various aspects of the validation process.
Returns:
(dict): Validation metrics obtained from the validation process.
Raises:
AssertionError: If the model is not a PyTorch model.
"""
custom = {"rect": True} # method defaults
args = {**self.overrides, **custom, **kwargs, "mode": "val"} # highest priority args on the right
validator = (validator or self._smart_load("validator"))(args=args, _callbacks=self.callbacks)
validator(model=self.model)
self.metrics = validator.metrics
return validator.metrics
def benchmark(
self,
**kwargs,
):
"""
Benchmarks the model across various export formats to evaluate performance.
This method assesses the model's performance in different export formats, such as ONNX, TorchScript, etc.
It uses the 'benchmark' function from the ultralytics.utils.benchmarks module. The benchmarking is configured
using a combination of default configuration values, model-specific arguments, method-specific defaults, and
any additional user-provided keyword arguments.
The method supports various arguments that allow customization of the benchmarking process, such as dataset
choice, image size, precision modes, device selection, and verbosity. For a comprehensive list of all
configurable options, users should refer to the 'configuration' section in the documentation.
Args:
**kwargs (any): Arbitrary keyword arguments to customize the benchmarking process. These are combined with
default configurations, model-specific arguments, and method defaults.
Returns:
(dict): A dictionary containing the results of the benchmarking process.
Raises:
AssertionError: If the model is not a PyTorch model.
"""
self._check_is_pytorch_model()
from ultralytics.utils.benchmarks import benchmark
custom = {"verbose": False} # method defaults
args = {**DEFAULT_CFG_DICT, **self.model.args, **custom, **kwargs, "mode": "benchmark"}
return benchmark(
model=self,
data=kwargs.get("data"), # if no 'data' argument passed set data=None for default datasets
imgsz=args["imgsz"],
half=args["half"],
int8=args["int8"],
device=args["device"],
verbose=kwargs.get("verbose"),
)
def export(
self,
**kwargs,
):
"""
Exports the model to a different format suitable for deployment.
This method facilitates the export of the model to various formats (e.g., ONNX, TorchScript) for deployment
purposes. It uses the 'Exporter' class for the export process, combining model-specific overrides, method
defaults, and any additional arguments provided. The combined arguments are used to configure export settings.
The method supports a wide range of arguments to customize the export process. For a comprehensive list of all
possible arguments, refer to the 'configuration' section in the documentation.
Args:
**kwargs (any): Arbitrary keyword arguments to customize the export process. These are combined with the
model's overrides and method defaults.
Returns:
(object): The exported model in the specified format, or an object related to the export process.
Raises:
AssertionError: If the model is not a PyTorch model.
"""
self._check_is_pytorch_model()
from .exporter import Exporter
custom = {"imgsz": self.model.args["imgsz"], "batch": 1, "data": None, "verbose": False} # method defaults
args = {**self.overrides, **custom, **kwargs, "mode": "export"} # highest priority args on the right
return Exporter(overrides=args, _callbacks=self.callbacks)(model=self.model)
def train(
self,
trainer=None,
**kwargs,
):
"""
Trains the model using the specified dataset and training configuration.
This method facilitates model training with a range of customizable settings and configurations. It supports
training with a custom trainer or the default training approach defined in the method. The method handles
different scenarios, such as resuming training from a checkpoint, integrating with Ultralytics HUB, and
updating model and configuration after training.
When using Ultralytics HUB, if the session already has a loaded model, the method prioritizes HUB training
arguments and issues a warning if local arguments are provided. It checks for pip updates and combines default
configurations, method-specific defaults, and user-provided arguments to configure the training process. After
training, it updates the model and its configurations, and optionally attaches metrics.
Args:
trainer (BaseTrainer, optional): An instance of a custom trainer class for training the model. If None, the
method uses a default trainer. Defaults to None.
**kwargs (any): Arbitrary keyword arguments representing the training configuration. These arguments are
used to customize various aspects of the training process.
Returns:
(dict | None): Training metrics if available and training is successful; otherwise, None.
Raises:
AssertionError: If the model is not a PyTorch model.
PermissionError: If there is a permission issue with the HUB session.
ModuleNotFoundError: If the HUB SDK is not installed.
"""
self._check_is_pytorch_model()
if hasattr(self.session, "model") and self.session.model.id: # Ultralytics HUB session with loaded model
if any(kwargs):
LOGGER.warning("WARNING ⚠️ using HUB training arguments, ignoring local training arguments.")
kwargs = self.session.train_args # overwrite kwargs
checks.check_pip_update_available()
overrides = yaml_load(checks.check_yaml(kwargs["cfg"])) if kwargs.get("cfg") else self.overrides
custom = {"data": DEFAULT_CFG_DICT["data"] or TASK2DATA[self.task]} # method defaults
args = {**overrides, **custom, **kwargs, "mode": "train"} # highest priority args on the right
if args.get("resume"):
args["resume"] = self.ckpt_path
self.trainer = (trainer or self._smart_load("trainer"))(overrides=args, _callbacks=self.callbacks)
if not args.get("resume"): # manually set model only if not resuming
self.trainer.model = self.trainer.get_model(weights=self.model if self.ckpt else None, cfg=self.model.yaml)
self.model = self.trainer.model
if SETTINGS["hub"] is True and not self.session:
# Create a model in HUB
try:
self.session = self._get_hub_session(self.model_name)
if self.session:
self.session.create_model(args)
# Check model was created
if not getattr(self.session.model, "id", None):
self.session = None
except (PermissionError, ModuleNotFoundError):
# Ignore PermissionError and ModuleNotFoundError which indicates hub-sdk not installed
pass
self.trainer.hub_session = self.session # attach optional HUB session
self.trainer.train()
# Update model and cfg after training
if RANK in (-1, 0):
ckpt = self.trainer.best if self.trainer.best.exists() else self.trainer.last
self.model, _ = attempt_load_one_weight(ckpt)
self.overrides = self.model.args
self.metrics = getattr(self.trainer.validator, "metrics", None) # TODO: no metrics returned by DDP
return self.metrics
def tune(
self,
use_ray=False,
iterations=10,
*args,
**kwargs,
):
"""
Conducts hyperparameter tuning for the model, with an option to use Ray Tune.
This method supports two modes of hyperparameter tuning: using Ray Tune or a custom tuning method.
When Ray Tune is enabled, it leverages the 'run_ray_tune' function from the ultralytics.utils.tuner module.
Otherwise, it uses the internal 'Tuner' class for tuning. The method combines default, overridden, and
custom arguments to configure the tuning process.
Args:
use_ray (bool): If True, uses Ray Tune for hyperparameter tuning. Defaults to False.
iterations (int): The number of tuning iterations to perform. Defaults to 10.
*args (list): Variable length argument list for additional arguments.
**kwargs (any): Arbitrary keyword arguments. These are combined with the model's overrides and defaults.
Returns:
(dict): A dictionary containing the results of the hyperparameter search.
Raises:
AssertionError: If the model is not a PyTorch model.
"""
self._check_is_pytorch_model()
if use_ray:
from ultralytics.utils.tuner import run_ray_tune
return run_ray_tune(self, max_samples=iterations, *args, **kwargs)
else:
from .tuner import Tuner
custom = {} # method defaults
args = {**self.overrides, **custom, **kwargs, "mode": "train"} # highest priority args on the right
return Tuner(args=args, _callbacks=self.callbacks)(model=self, iterations=iterations)
def _apply(self, fn) -> "Model":
"""Apply to(), cpu(), cuda(), half(), float() to model tensors that are not parameters or registered buffers."""
self._check_is_pytorch_model()
self = super()._apply(fn) # noqa
self.predictor = None # reset predictor as device may have changed
self.overrides["device"] = self.device # was str(self.device) i.e. device(type='cuda', index=0) -> 'cuda:0'
return self
@property
def names(self) -> list:
"""
Retrieves the class names associated with the loaded model.
This property returns the class names if they are defined in the model. It checks the class names for validity
using the 'check_class_names' function from the ultralytics.nn.autobackend module.
Returns:
(list | None): The class names of the model if available, otherwise None.
"""
from ultralytics.nn.autobackend import check_class_names
return check_class_names(self.model.names) if hasattr(self.model, "names") else None
@property
def device(self) -> torch.device:
"""
Retrieves the device on which the model's parameters are allocated.
This property is used to determine whether the model's parameters are on CPU or GPU. It only applies to models
that are instances of nn.Module.
Returns:
(torch.device | None): The device (CPU/GPU) of the model if it is a PyTorch model, otherwise None.
"""
return next(self.model.parameters()).device if isinstance(self.model, nn.Module) else None
@property
def transforms(self):
"""
Retrieves the transformations applied to the input data of the loaded model.
This property returns the transformations if they are defined in the model.
Returns:
(object | None): The transform object of the model if available, otherwise None.
"""
return self.model.transforms if hasattr(self.model, "transforms") else None
def add_callback(self, event: str, func) -> None:
"""
Adds a callback function for a specified event.
This method allows the user to register a custom callback function that is triggered on a specific event during
model training or inference.
Args:
event (str): The name of the event to attach the callback to.
func (callable): The callback function to be registered.
Raises:
ValueError: If the event name is not recognized.
"""
self.callbacks[event].append(func)
def clear_callback(self, event: str) -> None:
"""
Clears all callback functions registered for a specified event.
This method removes all custom and default callback functions associated with the given event.
Args:
event (str): The name of the event for which to clear the callbacks.
Raises:
ValueError: If the event name is not recognized.
"""
self.callbacks[event] = []
def reset_callbacks(self) -> None:
"""
Resets all callbacks to their default functions.
This method reinstates the default callback functions for all events, removing any custom callbacks that were
added previously.
"""
for event in callbacks.default_callbacks.keys():
self.callbacks[event] = [callbacks.default_callbacks[event][0]]
@staticmethod
def _reset_ckpt_args(args: dict) -> dict:
"""Reset arguments when loading a PyTorch model."""
include = {"imgsz", "data", "task", "single_cls"} # only remember these arguments when loading a PyTorch model
return {k: v for k, v in args.items() if k in include}
# def __getattr__(self, attr):
# """Raises error if object has no requested attribute."""
# name = self.__class__.__name__
# raise AttributeError(f"'{name}' object has no attribute '{attr}'. See valid attributes below.\n{self.__doc__}")
def _smart_load(self, key: str):
"""Load model/trainer/validator/predictor."""
try:
return self.task_map[self.task][key]
except Exception as e:
name = self.__class__.__name__
mode = inspect.stack()[1][3] # get the function name.
raise NotImplementedError(
emojis(f"WARNING ⚠️ '{name}' model does not support '{mode}' mode for '{self.task}' task yet.")
) from e
@property
def task_map(self) -> dict:
"""
Map head to model, trainer, validator, and predictor classes.
Returns:
task_map (dict): The map of model task to mode classes.
"""
raise NotImplementedError("Please provide task map for your model!")

397
ultralytics/engine/predictor.py
View File

@ -1,397 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
"""
Run prediction on images, videos, directories, globs, YouTube, webcam, streams, etc.
Usage - sources:
$ yolo mode=predict model=yolov8n.pt source=0 # webcam
img.jpg # image
vid.mp4 # video
screen # screenshot
path/ # directory
list.txt # list of images
list.streams # list of streams
'path/*.jpg' # glob
'https://youtu.be/LNwODJXcvt4' # YouTube
'rtsp://example.com/media.mp4' # RTSP, RTMP, HTTP, TCP stream
Usage - formats:
$ yolo mode=predict model=yolov8n.pt # PyTorch
yolov8n.torchscript # TorchScript
yolov8n.onnx # ONNX Runtime or OpenCV DNN with dnn=True
yolov8n_openvino_model # OpenVINO
yolov8n.engine # TensorRT
yolov8n.mlpackage # CoreML (macOS-only)
yolov8n_saved_model # TensorFlow SavedModel
yolov8n.pb # TensorFlow GraphDef
yolov8n.tflite # TensorFlow Lite
yolov8n_edgetpu.tflite # TensorFlow Edge TPU
yolov8n_paddle_model # PaddlePaddle
yolov8n_ncnn_model # NCNN
"""
import platform
import re
import threading
from pathlib import Path
import cv2
import numpy as np
import torch
from ultralytics.cfg import get_cfg, get_save_dir
from ultralytics.data import load_inference_source
from ultralytics.data.augment import LetterBox, classify_transforms
from ultralytics.nn.autobackend import AutoBackend
from ultralytics.utils import DEFAULT_CFG, LOGGER, MACOS, WINDOWS, callbacks, colorstr, ops
from ultralytics.utils.checks import check_imgsz, check_imshow
from ultralytics.utils.files import increment_path
from ultralytics.utils.torch_utils import select_device, smart_inference_mode
STREAM_WARNING = """
WARNING ⚠️ inference results will accumulate in RAM unless `stream=True` is passed, causing potential out-of-memory
errors for large sources or long-running streams and videos. See https://docs.ultralytics.com/modes/predict/ for help.
Example:
results = model(source=..., stream=True) # generator of Results objects
for r in results:
boxes = r.boxes # Boxes object for bbox outputs
masks = r.masks # Masks object for segment masks outputs
probs = r.probs # Class probabilities for classification outputs
"""
class BasePredictor:
"""
BasePredictor.
A base class for creating predictors.
Attributes:
args (SimpleNamespace): Configuration for the predictor.
save_dir (Path): Directory to save results.
done_warmup (bool): Whether the predictor has finished setup.
model (nn.Module): Model used for prediction.
data (dict): Data configuration.
device (torch.device): Device used for prediction.
dataset (Dataset): Dataset used for prediction.
vid_writer (dict): Dictionary of {save_path: video_writer, ...} writer for saving video output.
"""
def __init__(self, cfg=DEFAULT_CFG, overrides=None, _callbacks=None):
"""
Initializes the BasePredictor class.
Args:
cfg (str, optional): Path to a configuration file. Defaults to DEFAULT_CFG.
overrides (dict, optional): Configuration overrides. Defaults to None.
"""
self.args = get_cfg(cfg, overrides)
self.save_dir = get_save_dir(self.args)
if self.args.conf is None:
self.args.conf = 0.25 # default conf=0.25
self.done_warmup = False
if self.args.show:
self.args.show = check_imshow(warn=True)
# Usable if setup is done
self.model = None
self.data = self.args.data # data_dict
self.imgsz = None
self.device = None
self.dataset = None
self.vid_writer = {} # dict of {save_path: video_writer, ...}
self.plotted_img = None
self.source_type = None
self.seen = 0
self.windows = []
self.batch = None
self.results = None
self.transforms = None
self.callbacks = _callbacks or callbacks.get_default_callbacks()
self.txt_path = None
self._lock = threading.Lock() # for automatic thread-safe inference
callbacks.add_integration_callbacks(self)
def preprocess(self, im):
"""
Prepares input image before inference.
Args:
im (torch.Tensor | List(np.ndarray)): BCHW for tensor, [(HWC) x B] for list.
"""
not_tensor = not isinstance(im, torch.Tensor)
if not_tensor:
im = np.stack(self.pre_transform(im))
im = im[..., ::-1].transpose((0, 3, 1, 2)) # BGR to RGB, BHWC to BCHW, (n, 3, h, w)
im = np.ascontiguousarray(im) # contiguous
im = torch.from_numpy(im)
im = im.to(self.device)
im = im.half() if self.model.fp16 else im.float() # uint8 to fp16/32
if not_tensor:
im /= 255 # 0 - 255 to 0.0 - 1.0
return im
def inference(self, im, *args, **kwargs):
"""Runs inference on a given image using the specified model and arguments."""
visualize = (
increment_path(self.save_dir / Path(self.batch[0][0]).stem, mkdir=True)
if self.args.visualize and (not self.source_type.tensor)
else False
)
return self.model(im, augment=self.args.augment, visualize=visualize, embed=self.args.embed, *args, **kwargs)
def pre_transform(self, im):
"""
Pre-transform input image before inference.
Args:
im (List(np.ndarray)): (N, 3, h, w) for tensor, [(h, w, 3) x N] for list.
Returns:
(list): A list of transformed images.
"""
same_shapes = len({x.shape for x in im}) == 1
letterbox = LetterBox(self.imgsz, auto=same_shapes and self.model.pt, stride=self.model.stride)
return [letterbox(image=x) for x in im]
def postprocess(self, preds, img, orig_imgs):
"""Post-processes predictions for an image and returns them."""
return preds
def __call__(self, source=None, model=None, stream=False, *args, **kwargs):
"""Performs inference on an image or stream."""
self.stream = stream
if stream:
return self.stream_inference(source, model, *args, **kwargs)
else:
return list(self.stream_inference(source, model, *args, **kwargs)) # merge list of Result into one
def predict_cli(self, source=None, model=None):
"""
Method used for CLI prediction.
It uses always generator as outputs as not required by CLI mode.
"""
gen = self.stream_inference(source, model)
for _ in gen: # noqa, running CLI inference without accumulating any outputs (do not modify)
pass
def setup_source(self, source):
"""Sets up source and inference mode."""
self.imgsz = check_imgsz(self.args.imgsz, stride=self.model.stride, min_dim=2) # check image size
self.transforms = (
getattr(
self.model.model,
"transforms",
classify_transforms(self.imgsz[0], crop_fraction=self.args.crop_fraction),
)
if self.args.task == "classify"
else None
)
self.dataset = load_inference_source(
source=source,
batch=self.args.batch,
vid_stride=self.args.vid_stride,
buffer=self.args.stream_buffer,
)
self.source_type = self.dataset.source_type
if not getattr(self, "stream", True) and (
self.source_type.stream
or self.source_type.screenshot
or len(self.dataset) > 1000 # many images
or any(getattr(self.dataset, "video_flag", [False]))
): # videos
LOGGER.warning(STREAM_WARNING)
self.vid_writer = {}
@smart_inference_mode()
def stream_inference(self, source=None, model=None, *args, **kwargs):
"""Streams real-time inference on camera feed and saves results to file."""
if self.args.verbose:
LOGGER.info("")
# Setup model
if not self.model:
self.setup_model(model)
with self._lock: # for thread-safe inference
# Setup source every time predict is called
self.setup_source(source if source is not None else self.args.source)
# Check if save_dir/ label file exists
if self.args.save or self.args.save_txt:
(self.save_dir / "labels" if self.args.save_txt else self.save_dir).mkdir(parents=True, exist_ok=True)
# Warmup model
if not self.done_warmup:
self.model.warmup(imgsz=(1 if self.model.pt or self.model.triton else self.dataset.bs, 3, *self.imgsz))
self.done_warmup = True
self.seen, self.windows, self.batch = 0, [], None
profilers = (
ops.Profile(device=self.device),
ops.Profile(device=self.device),
ops.Profile(device=self.device),
)
self.run_callbacks("on_predict_start")
for self.batch in self.dataset:
self.run_callbacks("on_predict_batch_start")
paths, im0s, s = self.batch
# Preprocess
with profilers[0]:
im = self.preprocess(im0s)
# Inference
with profilers[1]:
preds = self.inference(im, *args, **kwargs)
if self.args.embed:
yield from [preds] if isinstance(preds, torch.Tensor) else preds # yield embedding tensors
continue
# Postprocess
with profilers[2]:
self.results = self.postprocess(preds, im, im0s)
self.run_callbacks("on_predict_postprocess_end")
# Visualize, save, write results
n = len(im0s)
for i in range(n):
self.seen += 1
self.results[i].speed = {
"preprocess": profilers[0].dt * 1e3 / n,
"inference": profilers[1].dt * 1e3 / n,
"postprocess": profilers[2].dt * 1e3 / n,
}
if self.args.verbose or self.args.save or self.args.save_txt or self.args.show:
s[i] += self.write_results(i, Path(paths[i]), im, s)
# Print batch results
if self.args.verbose:
LOGGER.info("\n".join(s))
self.run_callbacks("on_predict_batch_end")
yield from self.results
# Release assets
for v in self.vid_writer.values():
if isinstance(v, cv2.VideoWriter):
v.release()
# Print final results
if self.args.verbose and self.seen:
t = tuple(x.t / self.seen * 1e3 for x in profilers) # speeds per image
LOGGER.info(
f"Speed: %.1fms preprocess, %.1fms inference, %.1fms postprocess per image at shape "
f"{(min(self.args.batch, self.seen), 3, *im.shape[2:])}" % t
)
if self.args.save or self.args.save_txt or self.args.save_crop:
nl = len(list(self.save_dir.glob("labels/*.txt"))) # number of labels
s = f"\n{nl} label{'s' * (nl > 1)} saved to {self.save_dir / 'labels'}" if self.args.save_txt else ""
LOGGER.info(f"Results saved to {colorstr('bold', self.save_dir)}{s}")
self.run_callbacks("on_predict_end")
def setup_model(self, model, verbose=True):
"""Initialize YOLO model with given parameters and set it to evaluation mode."""
self.model = AutoBackend(
weights=model or self.args.model,
device=select_device(self.args.device, verbose=verbose),
dnn=self.args.dnn,
data=self.args.data,
fp16=self.args.half,
batch=self.args.batch,
fuse=True,
verbose=verbose,
)
self.device = self.model.device # update device
self.args.half = self.model.fp16 # update half
self.model.eval()
def write_results(self, i, p, im, s):
"""Write inference results to a file or directory."""
string = "" # print string
if len(im.shape) == 3:
im = im[None] # expand for batch dim
if self.source_type.stream or self.source_type.from_img or self.source_type.tensor: # batch_size >= 1
string += f"{i}: "
frame = self.dataset.count
else:
match = re.search(r"frame (\d+)/", s[i])
frame = int(match.group(1)) if match else None # 0 if frame undetermined
self.txt_path = self.save_dir / "labels" / (p.stem + ("" if self.dataset.mode == "image" else f"_{frame}"))
string += "%gx%g " % im.shape[2:]
result = self.results[i]
result.save_dir = self.save_dir.__str__() # used in other locations
string += result.verbose() + f"{result.speed['inference']:.1f}ms"
# Add predictions to image
if self.args.save or self.args.show:
self.plotted_img = result.plot(
line_width=self.args.line_width,
boxes=self.args.show_boxes,
conf=self.args.show_conf,
labels=self.args.show_labels,
im_gpu=None if self.args.retina_masks else im[i],
)
# Save results
if self.args.save_txt:
result.save_txt(f"{self.txt_path}.txt", save_conf=self.args.save_conf)
if self.args.save_crop:
result.save_crop(save_dir=self.save_dir / "crops", file_name=self.txt_path.stem)
if self.args.show:
self.show(str(p))
if self.args.save:
self.save_predicted_images(str(self.save_dir / (p.name or "tmp.jpg")), frame)
return string
def save_predicted_images(self, save_path="", frame=0):
"""Save video predictions as mp4 at specified path."""
im = self.plotted_img
# Save videos and streams
if self.dataset.mode in {"stream", "video"}:
fps = self.dataset.fps if self.dataset.mode == "video" else 30
frames_path = f'{save_path.split(".", 1)[0]}_frames/'
if save_path not in self.vid_writer: # new video
if self.args.save_frames:
Path(frames_path).mkdir(parents=True, exist_ok=True)
suffix, fourcc = (".mp4", "avc1") if MACOS else (".avi", "WMV2") if WINDOWS else (".avi", "MJPG")
self.vid_writer[save_path] = cv2.VideoWriter(
filename=str(Path(save_path).with_suffix(suffix)),
fourcc=cv2.VideoWriter_fourcc(*fourcc),
fps=fps, # integer required, floats produce error in MP4 codec
frameSize=(im.shape[1], im.shape[0]), # (width, height)
)
# Save video
self.vid_writer[save_path].write(im)
if self.args.save_frames:
cv2.imwrite(f"{frames_path}{frame}.jpg", im)
# Save images
else:
cv2.imwrite(save_path, im)
def show(self, p=""):
"""Display an image in a window using OpenCV imshow()."""
im = self.plotted_img
if platform.system() == "Linux" and p not in self.windows:
self.windows.append(p)
cv2.namedWindow(p, cv2.WINDOW_NORMAL | cv2.WINDOW_KEEPRATIO) # allow window resize (Linux)
cv2.resizeWindow(p, im.shape[1], im.shape[0]) # (width, height)
cv2.imshow(p, im)
cv2.waitKey(300 if self.dataset.mode == "image" else 1) # 1 millisecond
def run_callbacks(self, event: str):
"""Runs all registered callbacks for a specific event."""
for callback in self.callbacks.get(event, []):
callback(self)
def add_callback(self, event: str, func):
"""Add callback."""
self.callbacks[event].append(func)

743
ultralytics/engine/results.py
View File

@ -1,743 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
"""
Ultralytics Results, Boxes and Masks classes for handling inference results.
Usage: See https://docs.ultralytics.com/modes/predict/
"""
from copy import deepcopy
from functools import lru_cache
from pathlib import Path
import numpy as np
import torch
from ultralytics.data.augment import LetterBox
from ultralytics.utils import LOGGER, SimpleClass, ops
from ultralytics.utils.plotting import Annotator, colors, save_one_box
from ultralytics.utils.torch_utils import smart_inference_mode
class BaseTensor(SimpleClass):
"""Base tensor class with additional methods for easy manipulation and device handling."""
def __init__(self, data, orig_shape) -> None:
"""
Initialize BaseTensor with data and original shape.
Args:
data (torch.Tensor | np.ndarray): Predictions, such as bboxes, masks and keypoints.
orig_shape (tuple): Original shape of image.
"""
assert isinstance(data, (torch.Tensor, np.ndarray))
self.data = data
self.orig_shape = orig_shape
@property
def shape(self):
"""Return the shape of the data tensor."""
return self.data.shape
def cpu(self):
"""Return a copy of the tensor on CPU memory."""
return self if isinstance(self.data, np.ndarray) else self.__class__(self.data.cpu(), self.orig_shape)
def numpy(self):
"""Return a copy of the tensor as a numpy array."""
return self if isinstance(self.data, np.ndarray) else self.__class__(self.data.numpy(), self.orig_shape)
def cuda(self):
"""Return a copy of the tensor on GPU memory."""
return self.__class__(torch.as_tensor(self.data).cuda(), self.orig_shape)
def to(self, *args, **kwargs):
"""Return a copy of the tensor with the specified device and dtype."""
return self.__class__(torch.as_tensor(self.data).to(*args, **kwargs), self.orig_shape)
def __len__(self): # override len(results)
"""Return the length of the data tensor."""
return len(self.data)
def __getitem__(self, idx):
"""Return a BaseTensor with the specified index of the data tensor."""
return self.__class__(self.data[idx], self.orig_shape)
class Results(SimpleClass):
"""
A class for storing and manipulating inference results.
Attributes:
orig_img (numpy.ndarray): Original image as a numpy array.
orig_shape (tuple): Original image shape in (height, width) format.
boxes (Boxes, optional): Object containing detection bounding boxes.
masks (Masks, optional): Object containing detection masks.
probs (Probs, optional): Object containing class probabilities for classification tasks.
keypoints (Keypoints, optional): Object containing detected keypoints for each object.
speed (dict): Dictionary of preprocess, inference, and postprocess speeds (ms/image).
names (dict): Dictionary of class names.
path (str): Path to the image file.
Methods:
update(boxes=None, masks=None, probs=None, obb=None): Updates object attributes with new detection results.
cpu(): Returns a copy of the Results object with all tensors on CPU memory.
numpy(): Returns a copy of the Results object with all tensors as numpy arrays.
cuda(): Returns a copy of the Results object with all tensors on GPU memory.
to(*args, **kwargs): Returns a copy of the Results object with tensors on a specified device and dtype.
new(): Returns a new Results object with the same image, path, and names.
plot(...): Plots detection results on an input image, returning an annotated image.
show(): Show annotated results to screen.
save(filename): Save annotated results to file.
verbose(): Returns a log string for each task, detailing detections and classifications.
save_txt(txt_file, save_conf=False): Saves detection results to a text file.
save_crop(save_dir, file_name=Path("im.jpg")): Saves cropped detection images.
tojson(normalize=False): Converts detection results to JSON format.
"""
def __init__(self, orig_img, path, names, boxes=None, masks=None, probs=None, keypoints=None, obb=None) -> None:
"""
Initialize the Results class.
Args:
orig_img (numpy.ndarray): The original image as a numpy array.
path (str): The path to the image file.
names (dict): A dictionary of class names.
boxes (torch.tensor, optional): A 2D tensor of bounding box coordinates for each detection.
masks (torch.tensor, optional): A 3D tensor of detection masks, where each mask is a binary image.
probs (torch.tensor, optional): A 1D tensor of probabilities of each class for classification task.
keypoints (torch.tensor, optional): A 2D tensor of keypoint coordinates for each detection.
obb (torch.tensor, optional): A 2D tensor of oriented bounding box coordinates for each detection.
"""
self.orig_img = orig_img
self.orig_shape = orig_img.shape[:2]
self.boxes = Boxes(boxes, self.orig_shape) if boxes is not None else None # native size boxes
self.masks = Masks(masks, self.orig_shape) if masks is not None else None # native size or imgsz masks
self.probs = Probs(probs) if probs is not None else None
self.keypoints = Keypoints(keypoints, self.orig_shape) if keypoints is not None else None
self.obb = OBB(obb, self.orig_shape) if obb is not None else None
self.speed = {"preprocess": None, "inference": None, "postprocess": None} # milliseconds per image
self.names = names
self.path = path
self.save_dir = None
self._keys = "boxes", "masks", "probs", "keypoints", "obb"
def __getitem__(self, idx):
"""Return a Results object for the specified index."""
return self._apply("__getitem__", idx)
def __len__(self):
"""Return the number of detections in the Results object."""
for k in self._keys:
v = getattr(self, k)
if v is not None:
return len(v)
def update(self, boxes=None, masks=None, probs=None, obb=None):
"""Update the boxes, masks, and probs attributes of the Results object."""
if boxes is not None:
self.boxes = Boxes(ops.clip_boxes(boxes, self.orig_shape), self.orig_shape)
if masks is not None:
self.masks = Masks(masks, self.orig_shape)
if probs is not None:
self.probs = probs
if obb is not None:
self.obb = OBB(obb, self.orig_shape)
def _apply(self, fn, *args, **kwargs):
"""
Applies a function to all non-empty attributes and returns a new Results object with modified attributes. This
function is internally called by methods like .to(), .cuda(), .cpu(), etc.
Args:
fn (str): The name of the function to apply.
*args: Variable length argument list to pass to the function.
**kwargs: Arbitrary keyword arguments to pass to the function.
Returns:
Results: A new Results object with attributes modified by the applied function.
"""
r = self.new()
for k in self._keys:
v = getattr(self, k)
if v is not None:
setattr(r, k, getattr(v, fn)(*args, **kwargs))
return r
def cpu(self):
"""Return a copy of the Results object with all tensors on CPU memory."""
return self._apply("cpu")
def numpy(self):
"""Return a copy of the Results object with all tensors as numpy arrays."""
return self._apply("numpy")
def cuda(self):
"""Return a copy of the Results object with all tensors on GPU memory."""
return self._apply("cuda")
def to(self, *args, **kwargs):
"""Return a copy of the Results object with tensors on the specified device and dtype."""
return self._apply("to", *args, **kwargs)
def new(self):
"""Return a new Results object with the same image, path, and names."""
return Results(orig_img=self.orig_img, path=self.path, names=self.names)
def plot(
self,
conf=True,
line_width=None,
font_size=None,
font="Arial.ttf",
pil=False,
img=None,
im_gpu=None,
kpt_radius=5,
kpt_line=True,
labels=True,
boxes=True,
masks=True,
probs=True,
show=False,
save=False,
filename=None,
):
"""
Plots the detection results on an input RGB image. Accepts a numpy array (cv2) or a PIL Image.
Args:
conf (bool): Whether to plot the detection confidence score.
line_width (float, optional): The line width of the bounding boxes. If None, it is scaled to the image size.
font_size (float, optional): The font size of the text. If None, it is scaled to the image size.
font (str): The font to use for the text.
pil (bool): Whether to return the image as a PIL Image.
img (numpy.ndarray): Plot to another image. if not, plot to original image.
im_gpu (torch.Tensor): Normalized image in gpu with shape (1, 3, 640, 640), for faster mask plotting.
kpt_radius (int, optional): Radius of the drawn keypoints. Default is 5.
kpt_line (bool): Whether to draw lines connecting keypoints.
labels (bool): Whether to plot the label of bounding boxes.
boxes (bool): Whether to plot the bounding boxes.
masks (bool): Whether to plot the masks.
probs (bool): Whether to plot classification probability
show (bool): Whether to display the annotated image directly.
save (bool): Whether to save the annotated image to `filename`.
filename (str): Filename to save image to if save is True.
Returns:
(numpy.ndarray): A numpy array of the annotated image.
Example:
```python
from PIL import Image
from ultralytics import YOLO
model = YOLO('yolov8n.pt')
results = model('bus.jpg') # results list
for r in results:
im_array = r.plot() # plot a BGR numpy array of predictions
im = Image.fromarray(im_array[..., ::-1]) # RGB PIL image
im.show() # show image
im.save('results.jpg') # save image
```
"""
if img is None and isinstance(self.orig_img, torch.Tensor):
img = (self.orig_img[0].detach().permute(1, 2, 0).contiguous() * 255).to(torch.uint8).cpu().numpy()
names = self.names
is_obb = self.obb is not None
pred_boxes, show_boxes = self.obb if is_obb else self.boxes, boxes
pred_masks, show_masks = self.masks, masks
pred_probs, show_probs = self.probs, probs
annotator = Annotator(
deepcopy(self.orig_img if img is None else img),
line_width,
font_size,
font,
pil or (pred_probs is not None and show_probs), # Classify tasks default to pil=True
example=names,
)
# Plot Segment results
if pred_masks and show_masks:
if im_gpu is None:
img = LetterBox(pred_masks.shape[1:])(image=annotator.result())
im_gpu = (
torch.as_tensor(img, dtype=torch.float16, device=pred_masks.data.device)
.permute(2, 0, 1)
.flip(0)
.contiguous()
/ 255
)
idx = pred_boxes.cls if pred_boxes else range(len(pred_masks))
annotator.masks(pred_masks.data, colors=[colors(x, True) for x in idx], im_gpu=im_gpu)
# Plot Detect results
if pred_boxes is not None and show_boxes:
for d in reversed(pred_boxes):
c, conf, id = int(d.cls), float(d.conf) if conf else None, None if d.id is None else int(d.id.item())
name = ("" if id is None else f"id:{id} ") + names[c]
label = (f"{name} {conf:.2f}" if conf else name) if labels else None
box = d.xyxyxyxy.reshape(-1, 4, 2).squeeze() if is_obb else d.xyxy.squeeze()
annotator.box_label(box, label, color=colors(c, True), rotated=is_obb)
# Plot Classify results
if pred_probs is not None and show_probs:
text = ",\n".join(f"{names[j] if names else j} {pred_probs.data[j]:.2f}" for j in pred_probs.top5)
x = round(self.orig_shape[0] * 0.03)
annotator.text([x, x], text, txt_color=(255, 255, 255)) # TODO: allow setting colors
# Plot Pose results
if self.keypoints is not None:
for k in reversed(self.keypoints.data):
annotator.kpts(k, self.orig_shape, radius=kpt_radius, kpt_line=kpt_line)
# Show results
if show:
annotator.show(self.path)
# Save results
if save:
annotator.save(filename)
return annotator.result()
def show(self, *args, **kwargs):
"""Show annotated results image."""
self.plot(show=True, *args, **kwargs)
def save(self, filename=None, *args, **kwargs):
"""Save annotated results image."""
if not filename:
filename = f"results_{Path(self.path).name}"
self.plot(save=True, filename=filename, *args, **kwargs)
return filename
def verbose(self):
"""Return log string for each task."""
log_string = ""
probs = self.probs
boxes = self.boxes
if len(self) == 0:
return log_string if probs is not None else f"{log_string}(no detections), "
if probs is not None:
log_string += f"{', '.join(f'{self.names[j]} {probs.data[j]:.2f}' for j in probs.top5)}, "
if boxes:
for c in boxes.cls.unique():
n = (boxes.cls == c).sum() # detections per class
log_string += f"{n} {self.names[int(c)]}{'s' * (n > 1)}, "
return log_string
def save_txt(self, txt_file, save_conf=False):
"""
Save predictions into txt file.
Args:
txt_file (str): txt file path.
save_conf (bool): save confidence score or not.
"""
is_obb = self.obb is not None
boxes = self.obb if is_obb else self.boxes
masks = self.masks
probs = self.probs
kpts = self.keypoints
texts = []
if probs is not None:
# Classify
[texts.append(f"{probs.data[j]:.2f} {self.names[j]}") for j in probs.top5]
elif boxes:
# Detect/segment/pose
for j, d in enumerate(boxes):
c, conf, id = int(d.cls), float(d.conf), None if d.id is None else int(d.id.item())
line = (c, *(d.xyxyxyxyn.view(-1) if is_obb else d.xywhn.view(-1)))
if masks:
seg = masks[j].xyn[0].copy().reshape(-1) # reversed mask.xyn, (n,2) to (n*2)
line = (c, *seg)
if kpts is not None:
kpt = torch.cat((kpts[j].xyn, kpts[j].conf[..., None]), 2) if kpts[j].has_visible else kpts[j].xyn
line += (*kpt.reshape(-1).tolist(),)
line += (conf,) * save_conf + (() if id is None else (id,))
texts.append(("%g " * len(line)).rstrip() % line)
if texts:
Path(txt_file).parent.mkdir(parents=True, exist_ok=True) # make directory
with open(txt_file, "a") as f:
f.writelines(text + "\n" for text in texts)
def save_crop(self, save_dir, file_name=Path("im.jpg")):
"""
Save cropped predictions to `save_dir/cls/file_name.jpg`.
Args:
save_dir (str | pathlib.Path): Save path.
file_name (str | pathlib.Path): File name.
"""
if self.probs is not None:
LOGGER.warning("WARNING ⚠️ Classify task do not support `save_crop`.")
return
if self.obb is not None:
LOGGER.warning("WARNING ⚠️ OBB task do not support `save_crop`.")
return
for d in self.boxes:
save_one_box(
d.xyxy,
self.orig_img.copy(),
file=Path(save_dir) / self.names[int(d.cls)] / f"{Path(file_name)}.jpg",
BGR=True,
)
def summary(self, normalize=False, decimals=5):
"""Convert the results to a summarized format."""
if self.probs is not None:
LOGGER.warning("Warning: Classify results do not support the `summary()` method yet.")
return
# Create list of detection dictionaries
results = []
data = self.boxes.data.cpu().tolist()
h, w = self.orig_shape if normalize else (1, 1)
for i, row in enumerate(data): # xyxy, track_id if tracking, conf, class_id
box = {
"x1": round(row[0] / w, decimals),
"y1": round(row[1] / h, decimals),
"x2": round(row[2] / w, decimals),
"y2": round(row[3] / h, decimals),
}
conf = round(row[-2], decimals)
class_id = int(row[-1])
result = {"name": self.names[class_id], "class": class_id, "confidence": conf, "box": box}
if self.boxes.is_track:
result["track_id"] = int(row[-3]) # track ID
if self.masks:
result["segments"] = {
"x": (self.masks.xy[i][:, 0] / w).round(decimals).tolist(),
"y": (self.masks.xy[i][:, 1] / h).round(decimals).tolist(),
}
if self.keypoints is not None:
x, y, visible = self.keypoints[i].data[0].cpu().unbind(dim=1) # torch Tensor
result["keypoints"] = {
"x": (x / w).numpy().round(decimals).tolist(), # decimals named argument required
"y": (y / h).numpy().round(decimals).tolist(),
"visible": visible.numpy().round(decimals).tolist(),
}
results.append(result)
return results
def tojson(self, normalize=False, decimals=5):
"""Convert the results to JSON format."""
import json
return json.dumps(self.summary(normalize=normalize, decimals=decimals), indent=2)
class Boxes(BaseTensor):
"""
Manages detection boxes, providing easy access and manipulation of box coordinates, confidence scores, class
identifiers, and optional tracking IDs. Supports multiple formats for box coordinates, including both absolute and
normalized forms.
Attributes:
data (torch.Tensor): The raw tensor containing detection boxes and their associated data.
orig_shape (tuple): The original image size as a tuple (height, width), used for normalization.
is_track (bool): Indicates whether tracking IDs are included in the box data.
Properties:
xyxy (torch.Tensor | numpy.ndarray): Boxes in [x1, y1, x2, y2] format.
conf (torch.Tensor | numpy.ndarray): Confidence scores for each box.
cls (torch.Tensor | numpy.ndarray): Class labels for each box.
id (torch.Tensor | numpy.ndarray, optional): Tracking IDs for each box, if available.
xywh (torch.Tensor | numpy.ndarray): Boxes in [x, y, width, height] format, calculated on demand.
xyxyn (torch.Tensor | numpy.ndarray): Normalized [x1, y1, x2, y2] boxes, relative to `orig_shape`.
xywhn (torch.Tensor | numpy.ndarray): Normalized [x, y, width, height] boxes, relative to `orig_shape`.
Methods:
cpu(): Moves the boxes to CPU memory.
numpy(): Converts the boxes to a numpy array format.
cuda(): Moves the boxes to CUDA (GPU) memory.
to(device, dtype=None): Moves the boxes to the specified device.
"""
def __init__(self, boxes, orig_shape) -> None:
"""
Initialize the Boxes class.
Args:
boxes (torch.Tensor | numpy.ndarray): A tensor or numpy array containing the detection boxes, with
shape (num_boxes, 6) or (num_boxes, 7). The last two columns contain confidence and class values.
If present, the third last column contains track IDs.
orig_shape (tuple): Original image size, in the format (height, width).
"""
if boxes.ndim == 1:
boxes = boxes[None, :]
n = boxes.shape[-1]
assert n in (6, 7), f"expected 6 or 7 values but got {n}" # xyxy, track_id, conf, cls
super().__init__(boxes, orig_shape)
self.is_track = n == 7
self.orig_shape = orig_shape
@property
def xyxy(self):
"""Return the boxes in xyxy format."""
return self.data[:, :4]
@property
def conf(self):
"""Return the confidence values of the boxes."""
return self.data[:, -2]
@property
def cls(self):
"""Return the class values of the boxes."""
return self.data[:, -1]
@property
def id(self):
"""Return the track IDs of the boxes (if available)."""
return self.data[:, -3] if self.is_track else None
@property
@lru_cache(maxsize=2) # maxsize 1 should suffice
def xywh(self):
"""Return the boxes in xywh format."""
return ops.xyxy2xywh(self.xyxy)
@property
@lru_cache(maxsize=2)
def xyxyn(self):
"""Return the boxes in xyxy format normalized by original image size."""
xyxy = self.xyxy.clone() if isinstance(self.xyxy, torch.Tensor) else np.copy(self.xyxy)
xyxy[..., [0, 2]] /= self.orig_shape[1]
xyxy[..., [1, 3]] /= self.orig_shape[0]
return xyxy
@property
@lru_cache(maxsize=2)
def xywhn(self):
"""Return the boxes in xywh format normalized by original image size."""
xywh = ops.xyxy2xywh(self.xyxy)
xywh[..., [0, 2]] /= self.orig_shape[1]
xywh[..., [1, 3]] /= self.orig_shape[0]
return xywh
class Masks(BaseTensor):
"""
A class for storing and manipulating detection masks.
Attributes:
xy (list): A list of segments in pixel coordinates.
xyn (list): A list of normalized segments.
Methods:
cpu(): Returns the masks tensor on CPU memory.
numpy(): Returns the masks tensor as a numpy array.
cuda(): Returns the masks tensor on GPU memory.
to(device, dtype): Returns the masks tensor with the specified device and dtype.
"""
def __init__(self, masks, orig_shape) -> None:
"""Initialize the Masks class with the given masks tensor and original image shape."""
if masks.ndim == 2:
masks = masks[None, :]
super().__init__(masks, orig_shape)
@property
@lru_cache(maxsize=1)
def xyn(self):
"""Return normalized segments."""
return [
ops.scale_coords(self.data.shape[1:], x, self.orig_shape, normalize=True)
for x in ops.masks2segments(self.data)
]
@property
@lru_cache(maxsize=1)
def xy(self):
"""Return segments in pixel coordinates."""
return [
ops.scale_coords(self.data.shape[1:], x, self.orig_shape, normalize=False)
for x in ops.masks2segments(self.data)
]
class Keypoints(BaseTensor):
"""
A class for storing and manipulating detection keypoints.
Attributes:
xy (torch.Tensor): A collection of keypoints containing x, y coordinates for each detection.
xyn (torch.Tensor): A normalized version of xy with coordinates in the range [0, 1].
conf (torch.Tensor): Confidence values associated with keypoints if available, otherwise None.
Methods:
cpu(): Returns a copy of the keypoints tensor on CPU memory.
numpy(): Returns a copy of the keypoints tensor as a numpy array.
cuda(): Returns a copy of the keypoints tensor on GPU memory.
to(device, dtype): Returns a copy of the keypoints tensor with the specified device and dtype.
"""
@smart_inference_mode() # avoid keypoints < conf in-place error
def __init__(self, keypoints, orig_shape) -> None:
"""Initializes the Keypoints object with detection keypoints and original image size."""
if keypoints.ndim == 2:
keypoints = keypoints[None, :]
if keypoints.shape[2] == 3: # x, y, conf
mask = keypoints[..., 2] < 0.5 # points with conf < 0.5 (not visible)
keypoints[..., :2][mask] = 0
super().__init__(keypoints, orig_shape)
self.has_visible = self.data.shape[-1] == 3
@property
@lru_cache(maxsize=1)
def xy(self):
"""Returns x, y coordinates of keypoints."""
return self.data[..., :2]
@property
@lru_cache(maxsize=1)
def xyn(self):
"""Returns normalized x, y coordinates of keypoints."""
xy = self.xy.clone() if isinstance(self.xy, torch.Tensor) else np.copy(self.xy)
xy[..., 0] /= self.orig_shape[1]
xy[..., 1] /= self.orig_shape[0]
return xy
@property
@lru_cache(maxsize=1)
def conf(self):
"""Returns confidence values of keypoints if available, else None."""
return self.data[..., 2] if self.has_visible else None
class Probs(BaseTensor):
"""
A class for storing and manipulating classification predictions.
Attributes:
top1 (int): Index of the top 1 class.
top5 (list[int]): Indices of the top 5 classes.
top1conf (torch.Tensor): Confidence of the top 1 class.
top5conf (torch.Tensor): Confidences of the top 5 classes.
Methods:
cpu(): Returns a copy of the probs tensor on CPU memory.
numpy(): Returns a copy of the probs tensor as a numpy array.
cuda(): Returns a copy of the probs tensor on GPU memory.
to(): Returns a copy of the probs tensor with the specified device and dtype.
"""
def __init__(self, probs, orig_shape=None) -> None:
"""Initialize the Probs class with classification probabilities and optional original shape of the image."""
super().__init__(probs, orig_shape)
@property
@lru_cache(maxsize=1)
def top1(self):
"""Return the index of top 1."""
return int(self.data.argmax())
@property
@lru_cache(maxsize=1)
def top5(self):
"""Return the indices of top 5."""
return (-self.data).argsort(0)[:5].tolist() # this way works with both torch and numpy.
@property
@lru_cache(maxsize=1)
def top1conf(self):
"""Return the confidence of top 1."""
return self.data[self.top1]
@property
@lru_cache(maxsize=1)
def top5conf(self):
"""Return the confidences of top 5."""
return self.data[self.top5]
class OBB(BaseTensor):
"""
A class for storing and manipulating Oriented Bounding Boxes (OBB).
Args:
boxes (torch.Tensor | numpy.ndarray): A tensor or numpy array containing the detection boxes,
with shape (num_boxes, 7) or (num_boxes, 8). The last two columns contain confidence and class values.
If present, the third last column contains track IDs, and the fifth column from the left contains rotation.
orig_shape (tuple): Original image size, in the format (height, width).
Attributes:
xywhr (torch.Tensor | numpy.ndarray): The boxes in [x_center, y_center, width, height, rotation] format.
conf (torch.Tensor | numpy.ndarray): The confidence values of the boxes.
cls (torch.Tensor | numpy.ndarray): The class values of the boxes.
id (torch.Tensor | numpy.ndarray): The track IDs of the boxes (if available).
xyxyxyxyn (torch.Tensor | numpy.ndarray): The rotated boxes in xyxyxyxy format normalized by orig image size.
xyxyxyxy (torch.Tensor | numpy.ndarray): The rotated boxes in xyxyxyxy format.
xyxy (torch.Tensor | numpy.ndarray): The horizontal boxes in xyxyxyxy format.
data (torch.Tensor): The raw OBB tensor (alias for `boxes`).
Methods:
cpu(): Move the object to CPU memory.
numpy(): Convert the object to a numpy array.
cuda(): Move the object to CUDA memory.
to(*args, **kwargs): Move the object to the specified device.
"""
def __init__(self, boxes, orig_shape) -> None:
"""Initialize the Boxes class."""
if boxes.ndim == 1:
boxes = boxes[None, :]
n = boxes.shape[-1]
assert n in (7, 8), f"expected 7 or 8 values but got {n}" # xywh, rotation, track_id, conf, cls
super().__init__(boxes, orig_shape)
self.is_track = n == 8
self.orig_shape = orig_shape
@property
def xywhr(self):
"""Return the rotated boxes in xywhr format."""
return self.data[:, :5]
@property
def conf(self):
"""Return the confidence values of the boxes."""
return self.data[:, -2]
@property
def cls(self):
"""Return the class values of the boxes."""
return self.data[:, -1]
@property
def id(self):
"""Return the track IDs of the boxes (if available)."""
return self.data[:, -3] if self.is_track else None
@property
@lru_cache(maxsize=2)
def xyxyxyxy(self):
"""Return the boxes in xyxyxyxy format, (N, 4, 2)."""
return ops.xywhr2xyxyxyxy(self.xywhr)
@property
@lru_cache(maxsize=2)
def xyxyxyxyn(self):
"""Return the boxes in xyxyxyxy format, (N, 4, 2)."""
xyxyxyxyn = self.xyxyxyxy.clone() if isinstance(self.xyxyxyxy, torch.Tensor) else np.copy(self.xyxyxyxy)
xyxyxyxyn[..., 0] /= self.orig_shape[1]
xyxyxyxyn[..., 1] /= self.orig_shape[0]
return xyxyxyxyn
@property
@lru_cache(maxsize=2)
def xyxy(self):
"""
Return the horizontal boxes in xyxy format, (N, 4).
Accepts both torch and numpy boxes.
"""
x1 = self.xyxyxyxy[..., 0].min(1).values
x2 = self.xyxyxyxy[..., 0].max(1).values
y1 = self.xyxyxyxy[..., 1].min(1).values
y2 = self.xyxyxyxy[..., 1].max(1).values
xyxy = [x1, y1, x2, y2]
return np.stack(xyxy, axis=-1) if isinstance(self.data, np.ndarray) else torch.stack(xyxy, dim=-1)

757
ultralytics/engine/trainer.py
View File

@ -1,757 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
"""
Train a model on a dataset.
Usage:
$ yolo mode=train model=yolov8n.pt data=coco128.yaml imgsz=640 epochs=100 batch=16
"""
import math
import os
import subprocess
import time
import warnings
from copy import deepcopy
from datetime import datetime, timedelta
from pathlib import Path
import numpy as np
import torch
from torch import distributed as dist
from torch import nn, optim
from ultralytics.cfg import get_cfg, get_save_dir
from ultralytics.data.utils import check_cls_dataset, check_det_dataset
from ultralytics.nn.tasks import attempt_load_one_weight, attempt_load_weights
from ultralytics.utils import (
DEFAULT_CFG,
LOGGER,
RANK,
TQDM,
__version__,
callbacks,
clean_url,
colorstr,
emojis,
yaml_save,
)
from ultralytics.utils.autobatch import check_train_batch_size
from ultralytics.utils.checks import check_amp, check_file, check_imgsz, check_model_file_from_stem, print_args
from ultralytics.utils.dist import ddp_cleanup, generate_ddp_command
from ultralytics.utils.files import get_latest_run
from ultralytics.utils.torch_utils import (
EarlyStopping,
ModelEMA,
de_parallel,
init_seeds,
one_cycle,
select_device,
strip_optimizer,
)
class BaseTrainer:
"""
BaseTrainer.
A base class for creating trainers.
Attributes:
args (SimpleNamespace): Configuration for the trainer.
validator (BaseValidator): Validator instance.
model (nn.Module): Model instance.
callbacks (defaultdict): Dictionary of callbacks.
save_dir (Path): Directory to save results.
wdir (Path): Directory to save weights.
last (Path): Path to the last checkpoint.
best (Path): Path to the best checkpoint.
save_period (int): Save checkpoint every x epochs (disabled if < 1).
batch_size (int): Batch size for training.
epochs (int): Number of epochs to train for.
start_epoch (int): Starting epoch for training.
device (torch.device): Device to use for training.
amp (bool): Flag to enable AMP (Automatic Mixed Precision).
scaler (amp.GradScaler): Gradient scaler for AMP.
data (str): Path to data.
trainset (torch.utils.data.Dataset): Training dataset.
testset (torch.utils.data.Dataset): Testing dataset.
ema (nn.Module): EMA (Exponential Moving Average) of the model.
resume (bool): Resume training from a checkpoint.
lf (nn.Module): Loss function.
scheduler (torch.optim.lr_scheduler._LRScheduler): Learning rate scheduler.
best_fitness (float): The best fitness value achieved.
fitness (float): Current fitness value.
loss (float): Current loss value.
tloss (float): Total loss value.
loss_names (list): List of loss names.
csv (Path): Path to results CSV file.
"""
def __init__(self, cfg=DEFAULT_CFG, overrides=None, _callbacks=None):
"""
Initializes the BaseTrainer class.
Args:
cfg (str, optional): Path to a configuration file. Defaults to DEFAULT_CFG.
overrides (dict, optional): Configuration overrides. Defaults to None.
"""
self.args = get_cfg(cfg, overrides)
self.check_resume(overrides)
self.device = select_device(self.args.device, self.args.batch)
self.validator = None
self.metrics = None
self.plots = {}
init_seeds(self.args.seed + 1 + RANK, deterministic=self.args.deterministic)
# Dirs
self.save_dir = get_save_dir(self.args)
self.args.name = self.save_dir.name # update name for loggers
self.wdir = self.save_dir / "weights" # weights dir
if RANK in (-1, 0):
self.wdir.mkdir(parents=True, exist_ok=True) # make dir
self.args.save_dir = str(self.save_dir)
yaml_save(self.save_dir / "args.yaml", vars(self.args)) # save run args
self.last, self.best = self.wdir / "last.pt", self.wdir / "best.pt" # checkpoint paths
self.save_period = self.args.save_period
self.batch_size = self.args.batch
self.epochs = self.args.epochs
self.start_epoch = 0
if RANK == -1:
print_args(vars(self.args))
# Device
if self.device.type in ("cpu", "mps"):
self.args.workers = 0 # faster CPU training as time dominated by inference, not dataloading
# Model and Dataset
self.model = check_model_file_from_stem(self.args.model) # add suffix, i.e. yolov8n -> yolov8n.pt
try:
if self.args.task == "classify":
self.data = check_cls_dataset(self.args.data)
elif self.args.data.split(".")[-1] in ("yaml", "yml") or self.args.task in (
"detect",
"segment",
"pose",
"obb",
):
self.data = check_det_dataset(self.args.data)
if "yaml_file" in self.data:
self.args.data = self.data["yaml_file"] # for validating 'yolo train data=url.zip' usage
except Exception as e:
raise RuntimeError(emojis(f"Dataset '{clean_url(self.args.data)}' error ❌ {e}")) from e
self.trainset, self.testset = self.get_dataset(self.data)
self.ema = None
# Optimization utils init
self.lf = None
self.scheduler = None
# Epoch level metrics
self.best_fitness = None
self.fitness = None
self.loss = None
self.tloss = None
self.loss_names = ["Loss"]
self.csv = self.save_dir / "results.csv"
self.plot_idx = [0, 1, 2]
# Callbacks
self.callbacks = _callbacks or callbacks.get_default_callbacks()
if RANK in (-1, 0):
callbacks.add_integration_callbacks(self)
def add_callback(self, event: str, callback):
"""Appends the given callback."""
self.callbacks[event].append(callback)
def set_callback(self, event: str, callback):
"""Overrides the existing callbacks with the given callback."""
self.callbacks[event] = [callback]
def run_callbacks(self, event: str):
"""Run all existing callbacks associated with a particular event."""
for callback in self.callbacks.get(event, []):
callback(self)
def train(self):
"""Allow device='', device=None on Multi-GPU systems to default to device=0."""
if isinstance(self.args.device, str) and len(self.args.device): # i.e. device='0' or device='0,1,2,3'
world_size = len(self.args.device.split(","))
elif isinstance(self.args.device, (tuple, list)): # i.e. device=[0, 1, 2, 3] (multi-GPU from CLI is list)
world_size = len(self.args.device)
elif torch.cuda.is_available(): # i.e. device=None or device='' or device=number
world_size = 1 # default to device 0
else: # i.e. device='cpu' or 'mps'
world_size = 0
# Run subprocess if DDP training, else train normally
if world_size > 1 and "LOCAL_RANK" not in os.environ:
# Argument checks
if self.args.rect:
LOGGER.warning("WARNING ⚠️ 'rect=True' is incompatible with Multi-GPU training, setting 'rect=False'")
self.args.rect = False
if self.args.batch == -1:
LOGGER.warning(
"WARNING ⚠️ 'batch=-1' for AutoBatch is incompatible with Multi-GPU training, setting "
"default 'batch=16'"
)
self.args.batch = 16
# Command
cmd, file = generate_ddp_command(world_size, self)
try:
LOGGER.info(f'{colorstr("DDP:")} debug command {" ".join(cmd)}')
subprocess.run(cmd, check=True)
except Exception as e:
raise e
finally:
ddp_cleanup(self, str(file))
else:
self._do_train(world_size)
def _setup_scheduler(self):
"""Initialize training learning rate scheduler."""
if self.args.cos_lr:
self.lf = one_cycle(1, self.args.lrf, self.epochs) # cosine 1->hyp['lrf']
else:
self.lf = lambda x: max(1 - x / self.epochs, 0) * (1.0 - self.args.lrf) + self.args.lrf # linear
self.scheduler = optim.lr_scheduler.LambdaLR(self.optimizer, lr_lambda=self.lf)
def _setup_ddp(self, world_size):
"""Initializes and sets the DistributedDataParallel parameters for training."""
torch.cuda.set_device(RANK)
self.device = torch.device("cuda", RANK)
# LOGGER.info(f'DDP info: RANK {RANK}, WORLD_SIZE {world_size}, DEVICE {self.device}')
os.environ["NCCL_BLOCKING_WAIT"] = "1" # set to enforce timeout
dist.init_process_group(
backend="nccl" if dist.is_nccl_available() else "gloo",
timeout=timedelta(seconds=10800), # 3 hours
rank=RANK,
world_size=world_size,
)
def _setup_train(self, world_size):
"""Builds dataloaders and optimizer on correct rank process."""
# Model
self.run_callbacks("on_pretrain_routine_start")
ckpt = self.setup_model()
self.model = self.model.to(self.device)
self.set_model_attributes()
# Freeze layers
freeze_list = (
self.args.freeze
if isinstance(self.args.freeze, list)
else range(self.args.freeze)
if isinstance(self.args.freeze, int)
else []
)
always_freeze_names = [".dfl"] # always freeze these layers
freeze_layer_names = [f"model.{x}." for x in freeze_list] + always_freeze_names
for k, v in self.model.named_parameters():
# v.register_hook(lambda x: torch.nan_to_num(x)) # NaN to 0 (commented for erratic training results)
if any(x in k for x in freeze_layer_names):
LOGGER.info(f"Freezing layer '{k}'")
v.requires_grad = False
elif not v.requires_grad and v.dtype.is_floating_point: # only floating point Tensor can require gradients
LOGGER.info(
f"WARNING ⚠️ setting 'requires_grad=True' for frozen layer '{k}'. "
"See ultralytics.engine.trainer for customization of frozen layers."
)
v.requires_grad = True
# Check AMP
self.amp = torch.tensor(self.args.amp).to(self.device) # True or False
if self.amp and RANK in (-1, 0): # Single-GPU and DDP
callbacks_backup = callbacks.default_callbacks.copy() # backup callbacks as check_amp() resets them
self.amp = True# torch.tensor(check_amp(self.model), device=self.device)
callbacks.default_callbacks = callbacks_backup # restore callbacks
if RANK > -1 and world_size > 1: # DDP
dist.broadcast(self.amp, src=0) # broadcast the tensor from rank 0 to all other ranks (returns None)
self.amp = bool(self.amp) # as boolean
self.scaler = torch.cuda.amp.GradScaler(enabled=self.amp)
if world_size > 1:
self.model = nn.parallel.DistributedDataParallel(self.model, device_ids=[RANK])
# Check imgsz
gs = max(int(self.model.stride.max() if hasattr(self.model, "stride") else 32), 32) # grid size (max stride)
self.args.imgsz = check_imgsz(self.args.imgsz, stride=gs, floor=gs, max_dim=1)
self.stride = gs # for multiscale training
# Batch size
if self.batch_size == -1 and RANK == -1: # single-GPU only, estimate best batch size
self.args.batch = self.batch_size = check_train_batch_size(self.model, self.args.imgsz, self.amp)
# Dataloaders
batch_size = self.batch_size // max(world_size, 1)
self.train_loader = self.get_dataloader(self.trainset, batch_size=batch_size, rank=RANK, mode="train")
if RANK in (-1, 0):
# Note: When training DOTA dataset, double batch size could get OOM on images with >2000 objects.
self.test_loader = self.get_dataloader(
self.testset, batch_size=batch_size if self.args.task == "obb" else batch_size * 2, rank=-1, mode="val"
)
self.validator = self.get_validator()
metric_keys = self.validator.metrics.keys + self.label_loss_items(prefix="val")
self.metrics = dict(zip(metric_keys, [0] * len(metric_keys)))
self.ema = ModelEMA(self.model)
if self.args.plots:
self.plot_training_labels()
# Optimizer
self.accumulate = max(round(self.args.nbs / self.batch_size), 1) # accumulate loss before optimizing
weight_decay = self.args.weight_decay * self.batch_size * self.accumulate / self.args.nbs # scale weight_decay
iterations = math.ceil(len(self.train_loader.dataset) / max(self.batch_size, self.args.nbs)) * self.epochs
self.optimizer = self.build_optimizer(
model=self.model,
name=self.args.optimizer,
lr=self.args.lr0,
momentum=self.args.momentum,
decay=weight_decay,
iterations=iterations,
)
# Scheduler
self._setup_scheduler()
self.stopper, self.stop = EarlyStopping(patience=self.args.patience), False
self.resume_training(ckpt)
self.scheduler.last_epoch = self.start_epoch - 1 # do not move
self.run_callbacks("on_pretrain_routine_end")
def _do_train(self, world_size=1):
"""Train completed, evaluate and plot if specified by arguments."""
if world_size > 1:
self._setup_ddp(world_size)
self._setup_train(world_size)
nb = len(self.train_loader) # number of batches
nw = max(round(self.args.warmup_epochs * nb), 100) if self.args.warmup_epochs > 0 else -1 # warmup iterations
last_opt_step = -1
self.epoch_time = None
self.epoch_time_start = time.time()
self.train_time_start = time.time()
self.run_callbacks("on_train_start")
LOGGER.info(
f'Image sizes {self.args.imgsz} train, {self.args.imgsz} val\n'
f'Using {self.train_loader.num_workers * (world_size or 1)} dataloader workers\n'
f"Logging results to {colorstr('bold', self.save_dir)}\n"
f'Starting training for ' + (f"{self.args.time} hours..." if self.args.time else f"{self.epochs} epochs...")
)
if self.args.close_mosaic:
base_idx = (self.epochs - self.args.close_mosaic) * nb
self.plot_idx.extend([base_idx, base_idx + 1, base_idx + 2])
epoch = self.start_epoch
while True:
self.epoch = epoch
self.run_callbacks("on_train_epoch_start")
self.model.train()
if RANK != -1:
self.train_loader.sampler.set_epoch(epoch)
pbar = enumerate(self.train_loader)
# Update dataloader attributes (optional)
if epoch == (self.epochs - self.args.close_mosaic):
self._close_dataloader_mosaic()
self.train_loader.reset()
if RANK in (-1, 0):
LOGGER.info(self.progress_string())
pbar = TQDM(enumerate(self.train_loader), total=nb)
self.tloss = None
self.optimizer.zero_grad()
for i, batch in pbar:
self.run_callbacks("on_train_batch_start")
# Warmup
ni = i + nb * epoch
if ni <= nw:
xi = [0, nw] # x interp
self.accumulate = max(1, int(np.interp(ni, xi, [1, self.args.nbs / self.batch_size]).round()))
for j, x in enumerate(self.optimizer.param_groups):
# Bias lr falls from 0.1 to lr0, all other lrs rise from 0.0 to lr0
x["lr"] = np.interp(
ni, xi, [self.args.warmup_bias_lr if j == 0 else 0.0, x["initial_lr"] * self.lf(epoch)]
)
if "momentum" in x:
x["momentum"] = np.interp(ni, xi, [self.args.warmup_momentum, self.args.momentum])
# Forward
with torch.cuda.amp.autocast(self.amp):
batch = self.preprocess_batch(batch)
self.loss, self.loss_items = self.model(batch)
if RANK != -1:
self.loss *= world_size
self.tloss = (
(self.tloss * i + self.loss_items) / (i + 1) if self.tloss is not None else self.loss_items
)
# Backward
self.scaler.scale(self.loss).backward()
# Optimize - https://pytorch.org/docs/master/notes/amp_examples.html
if ni - last_opt_step >= self.accumulate:
self.optimizer_step()
last_opt_step = ni
# Timed stopping
if self.args.time:
self.stop = (time.time() - self.train_time_start) > (self.args.time * 3600)
if RANK != -1: # if DDP training
broadcast_list = [self.stop if RANK == 0 else None]
dist.broadcast_object_list(broadcast_list, 0) # broadcast 'stop' to all ranks
self.stop = broadcast_list[0]
if self.stop: # training time exceeded
break
# Log
mem = f"{torch.cuda.memory_reserved() / 1E9 if torch.cuda.is_available() else 0:.3g}G" # (GB)
loss_len = self.tloss.shape[0] if len(self.tloss.shape) else 1
losses = self.tloss if loss_len > 1 else torch.unsqueeze(self.tloss, 0)
if RANK in (-1, 0):
pbar.set_description(
("%11s" * 2 + "%11.4g" * (2 + loss_len))
% (f"{epoch + 1}/{self.epochs}", mem, *losses, batch["cls"].shape[0], batch["img"].shape[-1])
)
self.run_callbacks("on_batch_end")
if self.args.plots and ni in self.plot_idx:
self.plot_training_samples(batch, ni)
self.run_callbacks("on_train_batch_end")
self.lr = {f"lr/pg{ir}": x["lr"] for ir, x in enumerate(self.optimizer.param_groups)} # for loggers
self.run_callbacks("on_train_epoch_end")
# validate
if RANK in (-1, 0):
final_epoch = epoch + 1 == self.epochs
self.ema.update_attr(self.model, include=["yaml", "nc", "args", "names", "stride", "class_weights"])
# Validation
if (self.args.val and (((epoch+1) % self.args.val_period == 0) or (self.epochs - epoch) <= 10)) \
or final_epoch or self.stopper.possible_stop or self.stop:
self.metrics, self.fitness = self.validate()
self.save_metrics(metrics={**self.label_loss_items(self.tloss), **self.metrics, **self.lr})
self.stop |= self.stopper(epoch + 1, self.fitness) or final_epoch
if self.args.time:
self.stop |= (time.time() - self.train_time_start) > (self.args.time * 3600)
# Save model
if self.args.save or final_epoch:
self.save_model()
self.run_callbacks("on_model_save")
# Scheduler
t = time.time()
self.epoch_time = t - self.epoch_time_start
self.epoch_time_start = t
with warnings.catch_warnings():
warnings.simplefilter("ignore") # suppress 'Detected lr_scheduler.step() before optimizer.step()'
if self.args.time:
mean_epoch_time = (t - self.train_time_start) / (epoch - self.start_epoch + 1)
self.epochs = self.args.epochs = math.ceil(self.args.time * 3600 / mean_epoch_time)
self._setup_scheduler()
self.scheduler.last_epoch = self.epoch # do not move
self.stop |= epoch >= self.epochs # stop if exceeded epochs
self.scheduler.step()
self.run_callbacks("on_fit_epoch_end")
torch.cuda.empty_cache() # clear GPU memory at end of epoch, may help reduce CUDA out of memory errors
# Early Stopping
if RANK != -1: # if DDP training
broadcast_list = [self.stop if RANK == 0 else None]
dist.broadcast_object_list(broadcast_list, 0) # broadcast 'stop' to all ranks
self.stop = broadcast_list[0]
if self.stop:
break # must break all DDP ranks
epoch += 1
if RANK in (-1, 0):
# Do final val with best.pt
LOGGER.info(
f"\n{epoch - self.start_epoch + 1} epochs completed in "
f"{(time.time() - self.train_time_start) / 3600:.3f} hours."
)
self.final_eval()
if self.args.plots:
self.plot_metrics()
self.run_callbacks("on_train_end")
torch.cuda.empty_cache()
self.run_callbacks("teardown")
def save_model(self):
"""Save model training checkpoints with additional metadata."""
import pandas as pd # scope for faster startup
metrics = {**self.metrics, **{"fitness": self.fitness}}
results = {k.strip(): v for k, v in pd.read_csv(self.csv).to_dict(orient="list").items()}
ckpt = {
"epoch": self.epoch,
"best_fitness": self.best_fitness,
"model": deepcopy(de_parallel(self.model)).half(),
"ema": deepcopy(self.ema.ema).half(),
"updates": self.ema.updates,
"optimizer": self.optimizer.state_dict(),
"train_args": vars(self.args), # save as dict
"train_metrics": metrics,
"train_results": results,
"date": datetime.now().isoformat(),
"version": __version__,
"license": "AGPL-3.0 (https://ultralytics.com/license)",
"docs": "https://docs.ultralytics.com",
}
# Save last and best
torch.save(ckpt, self.last)
if self.best_fitness == self.fitness:
torch.save(ckpt, self.best)
if (self.save_period > 0) and (self.epoch > 0) and (self.epoch % self.save_period == 0):
torch.save(ckpt, self.wdir / f"epoch{self.epoch}.pt")
@staticmethod
def get_dataset(data):
"""
Get train, val path from data dict if it exists.
Returns None if data format is not recognized.
"""
return data["train"], data.get("val") or data.get("test")
def setup_model(self):
"""Load/create/download model for any task."""
if isinstance(self.model, torch.nn.Module): # if model is loaded beforehand. No setup needed
return
model, weights = self.model, None
ckpt = None
if str(model).endswith(".pt"):
weights, ckpt = attempt_load_one_weight(model)
cfg = ckpt["model"].yaml
else:
cfg = model
self.model = self.get_model(cfg=cfg, weights=weights, verbose=RANK == -1) # calls Model(cfg, weights)
return ckpt
def optimizer_step(self):
"""Perform a single step of the training optimizer with gradient clipping and EMA update."""
self.scaler.unscale_(self.optimizer) # unscale gradients
torch.nn.utils.clip_grad_norm_(self.model.parameters(), max_norm=10.0) # clip gradients
self.scaler.step(self.optimizer)
self.scaler.update()
self.optimizer.zero_grad()
if self.ema:
self.ema.update(self.model)
def preprocess_batch(self, batch):
"""Allows custom preprocessing model inputs and ground truths depending on task type."""
return batch
def validate(self):
"""
Runs validation on test set using self.validator.
The returned dict is expected to contain "fitness" key.
"""
metrics = self.validator(self)
fitness = metrics.pop("fitness", -self.loss.detach().cpu().numpy()) # use loss as fitness measure if not found
if not self.best_fitness or self.best_fitness < fitness:
self.best_fitness = fitness
return metrics, fitness
def get_model(self, cfg=None, weights=None, verbose=True):
"""Get model and raise NotImplementedError for loading cfg files."""
raise NotImplementedError("This task trainer doesn't support loading cfg files")
def get_validator(self):
"""Returns a NotImplementedError when the get_validator function is called."""
raise NotImplementedError("get_validator function not implemented in trainer")
def get_dataloader(self, dataset_path, batch_size=16, rank=0, mode="train"):
"""Returns dataloader derived from torch.data.Dataloader."""
raise NotImplementedError("get_dataloader function not implemented in trainer")
def build_dataset(self, img_path, mode="train", batch=None):
"""Build dataset."""
raise NotImplementedError("build_dataset function not implemented in trainer")
def label_loss_items(self, loss_items=None, prefix="train"):
"""
Returns a loss dict with labelled training loss items tensor.
Note:
This is not needed for classification but necessary for segmentation & detection
"""
return {"loss": loss_items} if loss_items is not None else ["loss"]
def set_model_attributes(self):
"""To set or update model parameters before training."""
self.model.names = self.data["names"]
def build_targets(self, preds, targets):
"""Builds target tensors for training YOLO model."""
pass
def progress_string(self):
"""Returns a string describing training progress."""
return ""
# TODO: may need to put these following functions into callback
def plot_training_samples(self, batch, ni):
"""Plots training samples during YOLO training."""
pass
def plot_training_labels(self):
"""Plots training labels for YOLO model."""
pass
def save_metrics(self, metrics):
"""Saves training metrics to a CSV file."""
keys, vals = list(metrics.keys()), list(metrics.values())
n = len(metrics) + 1 # number of cols
s = "" if self.csv.exists() else (("%23s," * n % tuple(["epoch"] + keys)).rstrip(",") + "\n") # header
with open(self.csv, "a") as f:
f.write(s + ("%23.5g," * n % tuple([self.epoch + 1] + vals)).rstrip(",") + "\n")
def plot_metrics(self):
"""Plot and display metrics visually."""
pass
def on_plot(self, name, data=None):
"""Registers plots (e.g. to be consumed in callbacks)"""
path = Path(name)
self.plots[path] = {"data": data, "timestamp": time.time()}
def final_eval(self):
"""Performs final evaluation and validation for object detection YOLO model."""
for f in self.last, self.best:
if f.exists():
strip_optimizer(f) # strip optimizers
if f is self.best:
LOGGER.info(f"\nValidating {f}...")
self.validator.args.plots = self.args.plots
self.metrics = self.validator(model=f)
self.metrics.pop("fitness", None)
self.run_callbacks("on_fit_epoch_end")
def check_resume(self, overrides):
"""Check if resume checkpoint exists and update arguments accordingly."""
resume = self.args.resume
if resume:
try:
exists = isinstance(resume, (str, Path)) and Path(resume).exists()
last = Path(check_file(resume) if exists else get_latest_run())
# Check that resume data YAML exists, otherwise strip to force re-download of dataset
ckpt_args = attempt_load_weights(last).args
if not Path(ckpt_args["data"]).exists():
ckpt_args["data"] = self.args.data
resume = True
self.args = get_cfg(ckpt_args)
self.args.model = self.args.resume = str(last) # reinstate model
for k in "imgsz", "batch", "device": # allow arg updates to reduce memory or update device on resume
if k in overrides:
setattr(self.args, k, overrides[k])
except Exception as e:
raise FileNotFoundError(
"Resume checkpoint not found. Please pass a valid checkpoint to resume from, "
"i.e. 'yolo train resume model=path/to/last.pt'"
) from e
self.resume = resume
def resume_training(self, ckpt):
"""Resume YOLO training from given epoch and best fitness."""
if ckpt is None or not self.resume:
return
best_fitness = 0.0
start_epoch = ckpt["epoch"] + 1
if ckpt["optimizer"] is not None:
self.optimizer.load_state_dict(ckpt["optimizer"]) # optimizer
best_fitness = ckpt["best_fitness"]
if self.ema and ckpt.get("ema"):
self.ema.ema.load_state_dict(ckpt["ema"].float().state_dict()) # EMA
self.ema.updates = ckpt["updates"]
assert start_epoch > 0, (
f"{self.args.model} training to {self.epochs} epochs is finished, nothing to resume.\n"
f"Start a new training without resuming, i.e. 'yolo train model={self.args.model}'"
)
LOGGER.info(f"Resuming training {self.args.model} from epoch {start_epoch + 1} to {self.epochs} total epochs")
if self.epochs < start_epoch:
LOGGER.info(
f"{self.model} has been trained for {ckpt['epoch']} epochs. Fine-tuning for {self.epochs} more epochs."
)
self.epochs += ckpt["epoch"] # finetune additional epochs
self.best_fitness = best_fitness
self.start_epoch = start_epoch
if start_epoch > (self.epochs - self.args.close_mosaic):
self._close_dataloader_mosaic()
def _close_dataloader_mosaic(self):
"""Update dataloaders to stop using mosaic augmentation."""
if hasattr(self.train_loader.dataset, "mosaic"):
self.train_loader.dataset.mosaic = False
if hasattr(self.train_loader.dataset, "close_mosaic"):
LOGGER.info("Closing dataloader mosaic")
self.train_loader.dataset.close_mosaic(hyp=self.args)
def build_optimizer(self, model, name="auto", lr=0.001, momentum=0.9, decay=1e-5, iterations=1e5):
"""
Constructs an optimizer for the given model, based on the specified optimizer name, learning rate, momentum,
weight decay, and number of iterations.
Args:
model (torch.nn.Module): The model for which to build an optimizer.
name (str, optional): The name of the optimizer to use. If 'auto', the optimizer is selected
based on the number of iterations. Default: 'auto'.
lr (float, optional): The learning rate for the optimizer. Default: 0.001.
momentum (float, optional): The momentum factor for the optimizer. Default: 0.9.
decay (float, optional): The weight decay for the optimizer. Default: 1e-5.
iterations (float, optional): The number of iterations, which determines the optimizer if
name is 'auto'. Default: 1e5.
Returns:
(torch.optim.Optimizer): The constructed optimizer.
"""
g = [], [], [] # optimizer parameter groups
bn = tuple(v for k, v in nn.__dict__.items() if "Norm" in k) # normalization layers, i.e. BatchNorm2d()
if name == "auto":
LOGGER.info(
f"{colorstr('optimizer:')} 'optimizer=auto' found, "
f"ignoring 'lr0={self.args.lr0}' and 'momentum={self.args.momentum}' and "
f"determining best 'optimizer', 'lr0' and 'momentum' automatically... "
)
nc = getattr(model, "nc", 10) # number of classes
lr_fit = round(0.002 * 5 / (4 + nc), 6) # lr0 fit equation to 6 decimal places
name, lr, momentum = ("SGD", 0.01, 0.9) if iterations > 10000 else ("AdamW", lr_fit, 0.9)
self.args.warmup_bias_lr = 0.0 # no higher than 0.01 for Adam
for module_name, module in model.named_modules():
for param_name, param in module.named_parameters(recurse=False):
fullname = f"{module_name}.{param_name}" if module_name else param_name
if "bias" in fullname: # bias (no decay)
g[2].append(param)
elif isinstance(module, bn): # weight (no decay)
g[1].append(param)
else: # weight (with decay)
g[0].append(param)
if name in ("Adam", "Adamax", "AdamW", "NAdam", "RAdam"):
optimizer = getattr(optim, name, optim.Adam)(g[2], lr=lr, betas=(momentum, 0.999), weight_decay=0.0)
elif name == "RMSProp":
optimizer = optim.RMSprop(g[2], lr=lr, momentum=momentum)
elif name == "SGD":
optimizer = optim.SGD(g[2], lr=lr, momentum=momentum, nesterov=True)
else:
raise NotImplementedError(
f"Optimizer '{name}' not found in list of available optimizers "
f"[Adam, AdamW, NAdam, RAdam, RMSProp, SGD, auto]."
"To request support for addition optimizers please visit https://github.com/ultralytics/ultralytics."
)
optimizer.add_param_group({"params": g[0], "weight_decay": decay}) # add g0 with weight_decay
optimizer.add_param_group({"params": g[1], "weight_decay": 0.0}) # add g1 (BatchNorm2d weights)
LOGGER.info(
f"{colorstr('optimizer:')} {type(optimizer).__name__}(lr={lr}, momentum={momentum}) with parameter groups "
f'{len(g[1])} weight(decay=0.0), {len(g[0])} weight(decay={decay}), {len(g[2])} bias(decay=0.0)'
)
return optimizer

242
ultralytics/engine/tuner.py
View File

@ -1,242 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
"""
This module provides functionalities for hyperparameter tuning of the Ultralytics YOLO models for object detection,
instance segmentation, image classification, pose estimation, and multi-object tracking.
Hyperparameter tuning is the process of systematically searching for the optimal set of hyperparameters
that yield the best model performance. This is particularly crucial in deep learning models like YOLO,
where small changes in hyperparameters can lead to significant differences in model accuracy and efficiency.
Example:
Tune hyperparameters for YOLOv8n on COCO8 at imgsz=640 and epochs=30 for 300 tuning iterations.
```python
from ultralytics import YOLO
model = YOLO('yolov8n.pt')
model.tune(data='coco8.yaml', epochs=10, iterations=300, optimizer='AdamW', plots=False, save=False, val=False)
```
"""
import random
import shutil
import subprocess
import time
import numpy as np
import torch
from ultralytics.cfg import get_cfg, get_save_dir
from ultralytics.utils import DEFAULT_CFG, LOGGER, callbacks, colorstr, remove_colorstr, yaml_print, yaml_save
from ultralytics.utils.plotting import plot_tune_results
class Tuner:
"""
Class responsible for hyperparameter tuning of YOLO models.
The class evolves YOLO model hyperparameters over a given number of iterations
by mutating them according to the search space and retraining the model to evaluate their performance.
Attributes:
space (dict): Hyperparameter search space containing bounds and scaling factors for mutation.
tune_dir (Path): Directory where evolution logs and results will be saved.
tune_csv (Path): Path to the CSV file where evolution logs are saved.
Methods:
_mutate(hyp: dict) -> dict:
Mutates the given hyperparameters within the bounds specified in `self.space`.
__call__():
Executes the hyperparameter evolution across multiple iterations.
Example:
Tune hyperparameters for YOLOv8n on COCO8 at imgsz=640 and epochs=30 for 300 tuning iterations.
```python
from ultralytics import YOLO
model = YOLO('yolov8n.pt')
model.tune(data='coco8.yaml', epochs=10, iterations=300, optimizer='AdamW', plots=False, save=False, val=False)
```
Tune with custom search space.
```python
from ultralytics import YOLO
model = YOLO('yolov8n.pt')
model.tune(space={key1: val1, key2: val2}) # custom search space dictionary
```
"""
def __init__(self, args=DEFAULT_CFG, _callbacks=None):
"""
Initialize the Tuner with configurations.
Args:
args (dict, optional): Configuration for hyperparameter evolution.
"""
self.space = args.pop("space", None) or { # key: (min, max, gain(optional))
# 'optimizer': tune.choice(['SGD', 'Adam', 'AdamW', 'NAdam', 'RAdam', 'RMSProp']),
"lr0": (1e-5, 1e-1), # initial learning rate (i.e. SGD=1E-2, Adam=1E-3)
"lrf": (0.0001, 0.1), # final OneCycleLR learning rate (lr0 * lrf)
"momentum": (0.7, 0.98, 0.3), # SGD momentum/Adam beta1
"weight_decay": (0.0, 0.001), # optimizer weight decay 5e-4
"warmup_epochs": (0.0, 5.0), # warmup epochs (fractions ok)
"warmup_momentum": (0.0, 0.95), # warmup initial momentum
"box": (1.0, 20.0), # box loss gain
"cls": (0.2, 4.0), # cls loss gain (scale with pixels)
"dfl": (0.4, 6.0), # dfl loss gain
"hsv_h": (0.0, 0.1), # image HSV-Hue augmentation (fraction)
"hsv_s": (0.0, 0.9), # image HSV-Saturation augmentation (fraction)
"hsv_v": (0.0, 0.9), # image HSV-Value augmentation (fraction)
"degrees": (0.0, 45.0), # image rotation (+/- deg)
"translate": (0.0, 0.9), # image translation (+/- fraction)
"scale": (0.0, 0.95), # image scale (+/- gain)
"shear": (0.0, 10.0), # image shear (+/- deg)
"perspective": (0.0, 0.001), # image perspective (+/- fraction), range 0-0.001
"flipud": (0.0, 1.0), # image flip up-down (probability)
"fliplr": (0.0, 1.0), # image flip left-right (probability)
"bgr": (0.0, 1.0), # image channel bgr (probability)
"mosaic": (0.0, 1.0), # image mixup (probability)
"mixup": (0.0, 1.0), # image mixup (probability)
"copy_paste": (0.0, 1.0), # segment copy-paste (probability)
}
self.args = get_cfg(overrides=args)
self.tune_dir = get_save_dir(self.args, name="tune")
self.tune_csv = self.tune_dir / "tune_results.csv"
self.callbacks = _callbacks or callbacks.get_default_callbacks()
self.prefix = colorstr("Tuner: ")
callbacks.add_integration_callbacks(self)
LOGGER.info(
f"{self.prefix}Initialized Tuner instance with 'tune_dir={self.tune_dir}'\n"
f"{self.prefix}💡 Learn about tuning at https://docs.ultralytics.com/guides/hyperparameter-tuning"
)
def _mutate(self, parent="single", n=5, mutation=0.8, sigma=0.2):
"""
Mutates the hyperparameters based on bounds and scaling factors specified in `self.space`.
Args:
parent (str): Parent selection method: 'single' or 'weighted'.
n (int): Number of parents to consider.
mutation (float): Probability of a parameter mutation in any given iteration.
sigma (float): Standard deviation for Gaussian random number generator.
Returns:
(dict): A dictionary containing mutated hyperparameters.
"""
if self.tune_csv.exists(): # if CSV file exists: select best hyps and mutate
# Select parent(s)
x = np.loadtxt(self.tune_csv, ndmin=2, delimiter=",", skiprows=1)
fitness = x[:, 0] # first column
n = min(n, len(x)) # number of previous results to consider
x = x[np.argsort(-fitness)][:n] # top n mutations
w = x[:, 0] - x[:, 0].min() + 1e-6 # weights (sum > 0)
if parent == "single" or len(x) == 1:
# x = x[random.randint(0, n - 1)] # random selection
x = x[random.choices(range(n), weights=w)[0]] # weighted selection
elif parent == "weighted":
x = (x * w.reshape(n, 1)).sum(0) / w.sum() # weighted combination
# Mutate
r = np.random # method
r.seed(int(time.time()))
g = np.array([v[2] if len(v) == 3 else 1.0 for k, v in self.space.items()]) # gains 0-1
ng = len(self.space)
v = np.ones(ng)
while all(v == 1): # mutate until a change occurs (prevent duplicates)
v = (g * (r.random(ng) < mutation) * r.randn(ng) * r.random() * sigma + 1).clip(0.3, 3.0)
hyp = {k: float(x[i + 1] * v[i]) for i, k in enumerate(self.space.keys())}
else:
hyp = {k: getattr(self.args, k) for k in self.space.keys()}
# Constrain to limits
for k, v in self.space.items():
hyp[k] = max(hyp[k], v[0]) # lower limit
hyp[k] = min(hyp[k], v[1]) # upper limit
hyp[k] = round(hyp[k], 5) # significant digits
return hyp
def __call__(self, model=None, iterations=10, cleanup=True):
"""
Executes the hyperparameter evolution process when the Tuner instance is called.
This method iterates through the number of iterations, performing the following steps in each iteration:
1. Load the existing hyperparameters or initialize new ones.
2. Mutate the hyperparameters using the `mutate` method.
3. Train a YOLO model with the mutated hyperparameters.
4. Log the fitness score and mutated hyperparameters to a CSV file.
Args:
model (Model): A pre-initialized YOLO model to be used for training.
iterations (int): The number of generations to run the evolution for.
cleanup (bool): Whether to delete iteration weights to reduce storage space used during tuning.
Note:
The method utilizes the `self.tune_csv` Path object to read and log hyperparameters and fitness scores.
Ensure this path is set correctly in the Tuner instance.
"""
t0 = time.time()
best_save_dir, best_metrics = None, None
(self.tune_dir / "weights").mkdir(parents=True, exist_ok=True)
for i in range(iterations):
# Mutate hyperparameters
mutated_hyp = self._mutate()
LOGGER.info(f"{self.prefix}Starting iteration {i + 1}/{iterations} with hyperparameters: {mutated_hyp}")
metrics = {}
train_args = {**vars(self.args), **mutated_hyp}
save_dir = get_save_dir(get_cfg(train_args))
weights_dir = save_dir / "weights"
try:
# Train YOLO model with mutated hyperparameters (run in subprocess to avoid dataloader hang)
cmd = ["yolo", "train", *(f"{k}={v}" for k, v in train_args.items())]
return_code = subprocess.run(cmd, check=True).returncode
ckpt_file = weights_dir / ("best.pt" if (weights_dir / "best.pt").exists() else "last.pt")
metrics = torch.load(ckpt_file)["train_metrics"]
assert return_code == 0, "training failed"
except Exception as e:
LOGGER.warning(f"WARNING ❌️ training failure for hyperparameter tuning iteration {i + 1}\n{e}")
# Save results and mutated_hyp to CSV
fitness = metrics.get("fitness", 0.0)
log_row = [round(fitness, 5)] + [mutated_hyp[k] for k in self.space.keys()]
headers = "" if self.tune_csv.exists() else (",".join(["fitness"] + list(self.space.keys())) + "\n")
with open(self.tune_csv, "a") as f:
f.write(headers + ",".join(map(str, log_row)) + "\n")
# Get best results
x = np.loadtxt(self.tune_csv, ndmin=2, delimiter=",", skiprows=1)
fitness = x[:, 0] # first column
best_idx = fitness.argmax()
best_is_current = best_idx == i
if best_is_current:
best_save_dir = save_dir
best_metrics = {k: round(v, 5) for k, v in metrics.items()}
for ckpt in weights_dir.glob("*.pt"):
shutil.copy2(ckpt, self.tune_dir / "weights")
elif cleanup:
shutil.rmtree(ckpt_file.parent) # remove iteration weights/ dir to reduce storage space
# Plot tune results
plot_tune_results(self.tune_csv)
# Save and print tune results
header = (
f'{self.prefix}{i + 1}/{iterations} iterations complete ✅ ({time.time() - t0:.2f}s)\n'
f'{self.prefix}Results saved to {colorstr("bold", self.tune_dir)}\n'
f'{self.prefix}Best fitness={fitness[best_idx]} observed at iteration {best_idx + 1}\n'
f'{self.prefix}Best fitness metrics are {best_metrics}\n'
f'{self.prefix}Best fitness model is {best_save_dir}\n'
f'{self.prefix}Best fitness hyperparameters are printed below.\n'
)
LOGGER.info("\n" + header)
data = {k: float(x[best_idx, i + 1]) for i, k in enumerate(self.space.keys())}
yaml_save(
self.tune_dir / "best_hyperparameters.yaml",
data=data,
header=remove_colorstr(header.replace(self.prefix, "# ")) + "\n",
)
yaml_print(self.tune_dir / "best_hyperparameters.yaml")

345
ultralytics/engine/validator.py
View File

@ -1,345 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
"""
Check a model's accuracy on a test or val split of a dataset.
Usage:
$ yolo mode=val model=yolov8n.pt data=coco128.yaml imgsz=640
Usage - formats:
$ yolo mode=val model=yolov8n.pt # PyTorch
yolov8n.torchscript # TorchScript
yolov8n.onnx # ONNX Runtime or OpenCV DNN with dnn=True
yolov8n_openvino_model # OpenVINO
yolov8n.engine # TensorRT
yolov8n.mlpackage # CoreML (macOS-only)
yolov8n_saved_model # TensorFlow SavedModel
yolov8n.pb # TensorFlow GraphDef
yolov8n.tflite # TensorFlow Lite
yolov8n_edgetpu.tflite # TensorFlow Edge TPU
yolov8n_paddle_model # PaddlePaddle
yolov8n_ncnn_model # NCNN
"""
import json
import time
from pathlib import Path
import numpy as np
import torch
from ultralytics.cfg import get_cfg, get_save_dir
from ultralytics.data.utils import check_cls_dataset, check_det_dataset
from ultralytics.nn.autobackend import AutoBackend
from ultralytics.utils import LOGGER, TQDM, callbacks, colorstr, emojis
from ultralytics.utils.checks import check_imgsz
from ultralytics.utils.ops import Profile
from ultralytics.utils.torch_utils import de_parallel, select_device, smart_inference_mode
class BaseValidator:
"""
BaseValidator.
A base class for creating validators.
Attributes:
args (SimpleNamespace): Configuration for the validator.
dataloader (DataLoader): Dataloader to use for validation.
pbar (tqdm): Progress bar to update during validation.
model (nn.Module): Model to validate.
data (dict): Data dictionary.
device (torch.device): Device to use for validation.
batch_i (int): Current batch index.
training (bool): Whether the model is in training mode.
names (dict): Class names.
seen: Records the number of images seen so far during validation.
stats: Placeholder for statistics during validation.
confusion_matrix: Placeholder for a confusion matrix.
nc: Number of classes.
iouv: (torch.Tensor): IoU thresholds from 0.50 to 0.95 in spaces of 0.05.
jdict (dict): Dictionary to store JSON validation results.
speed (dict): Dictionary with keys 'preprocess', 'inference', 'loss', 'postprocess' and their respective
batch processing times in milliseconds.
save_dir (Path): Directory to save results.
plots (dict): Dictionary to store plots for visualization.
callbacks (dict): Dictionary to store various callback functions.
"""
def __init__(self, dataloader=None, save_dir=None, pbar=None, args=None, _callbacks=None):
"""
Initializes a BaseValidator instance.
Args:
dataloader (torch.utils.data.DataLoader): Dataloader to be used for validation.
save_dir (Path, optional): Directory to save results.
pbar (tqdm.tqdm): Progress bar for displaying progress.
args (SimpleNamespace): Configuration for the validator.
_callbacks (dict): Dictionary to store various callback functions.
"""
self.args = get_cfg(overrides=args)
self.dataloader = dataloader
self.pbar = pbar
self.stride = None
self.data = None
self.device = None
self.batch_i = None
self.training = True
self.names = None
self.seen = None
self.stats = None
self.confusion_matrix = None
self.nc = None
self.iouv = None
self.jdict = None
self.speed = {"preprocess": 0.0, "inference": 0.0, "loss": 0.0, "postprocess": 0.0}
self.save_dir = save_dir or get_save_dir(self.args)
(self.save_dir / "labels" if self.args.save_txt else self.save_dir).mkdir(parents=True, exist_ok=True)
if self.args.conf is None:
self.args.conf = 0.001 # default conf=0.001
self.args.imgsz = check_imgsz(self.args.imgsz, max_dim=1)
self.plots = {}
self.callbacks = _callbacks or callbacks.get_default_callbacks()
@smart_inference_mode()
def __call__(self, trainer=None, model=None):
"""Supports validation of a pre-trained model if passed or a model being trained if trainer is passed (trainer
gets priority).
"""
self.training = trainer is not None
augment = self.args.augment and (not self.training)
if self.training:
self.device = trainer.device
self.data = trainer.data
# self.args.half = self.device.type != "cpu" # force FP16 val during training
model = trainer.ema.ema or trainer.model
model = model.half() if self.args.half else model.float()
# self.model = model
self.loss = torch.zeros_like(trainer.loss_items, device=trainer.device)
self.args.plots &= trainer.stopper.possible_stop or (trainer.epoch == trainer.epochs - 1)
model.eval()
else:
callbacks.add_integration_callbacks(self)
model = AutoBackend(
weights=model or self.args.model,
device=select_device(self.args.device, self.args.batch),
dnn=self.args.dnn,
data=self.args.data,
fp16=self.args.half,
)
# self.model = model
self.device = model.device # update device
self.args.half = model.fp16 # update half
stride, pt, jit, engine = model.stride, model.pt, model.jit, model.engine
imgsz = check_imgsz(self.args.imgsz, stride=stride)
if engine:
self.args.batch = model.batch_size
elif not pt and not jit:
self.args.batch = 1 # export.py models default to batch-size 1
LOGGER.info(f"Forcing batch=1 square inference (1,3,{imgsz},{imgsz}) for non-PyTorch models")
if str(self.args.data).split(".")[-1] in ("yaml", "yml"):
self.data = check_det_dataset(self.args.data)
elif self.args.task == "classify":
self.data = check_cls_dataset(self.args.data, split=self.args.split)
else:
raise FileNotFoundError(emojis(f"Dataset '{self.args.data}' for task={self.args.task} not found ❌"))
if self.device.type in ("cpu", "mps"):
self.args.workers = 0 # faster CPU val as time dominated by inference, not dataloading
if not pt:
self.args.rect = False
self.stride = model.stride # used in get_dataloader() for padding
self.dataloader = self.dataloader or self.get_dataloader(self.data.get(self.args.split), self.args.batch)
model.eval()
model.warmup(imgsz=(1 if pt else self.args.batch, 3, imgsz, imgsz)) # warmup
self.run_callbacks("on_val_start")
dt = (
Profile(device=self.device),
Profile(device=self.device),
Profile(device=self.device),
Profile(device=self.device),
)
bar = TQDM(self.dataloader, desc=self.get_desc(), total=len(self.dataloader))
self.init_metrics(de_parallel(model))
self.jdict = [] # empty before each val
for batch_i, batch in enumerate(bar):
self.run_callbacks("on_val_batch_start")
self.batch_i = batch_i
# Preprocess
with dt[0]:
batch = self.preprocess(batch)
# Inference
with dt[1]:
preds = model(batch["img"], augment=augment)
# Loss
with dt[2]:
if self.training:
self.loss += model.loss(batch, preds)[1]
# Postprocess
with dt[3]:
preds = self.postprocess(preds)
self.update_metrics(preds, batch)
if False:#self.args.plots and batch_i < 3:
self.plot_val_samples(batch, batch_i)
self.plot_predictions(batch, preds, batch_i)
self.run_callbacks("on_val_batch_end")
stats = self.get_stats()
self.check_stats(stats)
self.speed = dict(zip(self.speed.keys(), (x.t / len(self.dataloader.dataset) * 1e3 for x in dt)))
self.finalize_metrics()
if not (self.args.save_json and self.is_coco and len(self.jdict)):
self.print_results()
self.run_callbacks("on_val_end")
if self.training:
model.float()
if self.args.save_json and self.jdict:
with open(str(self.save_dir / "predictions.json"), "w") as f:
LOGGER.info(f"Saving {f.name}...")
json.dump(self.jdict, f) # flatten and save
stats = self.eval_json(stats) # update stats
stats['fitness'] = stats['metrics/mAP50-95(B)']
results = {**stats, **trainer.label_loss_items(self.loss.cpu() / len(self.dataloader), prefix="val")}
return {k: round(float(v), 5) for k, v in results.items()} # return results as 5 decimal place floats
else:
LOGGER.info(
"Speed: %.1fms preprocess, %.1fms inference, %.1fms loss, %.1fms postprocess per image"
% tuple(self.speed.values())
)
if self.args.save_json and self.jdict:
with open(str(self.save_dir / "predictions.json"), "w") as f:
LOGGER.info(f"Saving {f.name}...")
json.dump(self.jdict, f) # flatten and save
stats = self.eval_json(stats) # update stats
if self.args.plots or self.args.save_json:
LOGGER.info(f"Results saved to {colorstr('bold', self.save_dir)}")
return stats
def match_predictions(self, pred_classes, true_classes, iou, use_scipy=False):
"""
Matches predictions to ground truth objects (pred_classes, true_classes) using IoU.
Args:
pred_classes (torch.Tensor): Predicted class indices of shape(N,).
true_classes (torch.Tensor): Target class indices of shape(M,).
iou (torch.Tensor): An NxM tensor containing the pairwise IoU values for predictions and ground of truth
use_scipy (bool): Whether to use scipy for matching (more precise).
Returns:
(torch.Tensor): Correct tensor of shape(N,10) for 10 IoU thresholds.
"""
# Dx10 matrix, where D - detections, 10 - IoU thresholds
correct = np.zeros((pred_classes.shape[0], self.iouv.shape[0])).astype(bool)
# LxD matrix where L - labels (rows), D - detections (columns)
correct_class = true_classes[:, None] == pred_classes
iou = iou * correct_class # zero out the wrong classes
iou = iou.cpu().numpy()
for i, threshold in enumerate(self.iouv.cpu().tolist()):
if use_scipy:
# WARNING: known issue that reduces mAP in https://github.com/ultralytics/ultralytics/pull/4708
import scipy # scope import to avoid importing for all commands
cost_matrix = iou * (iou >= threshold)
if cost_matrix.any():
labels_idx, detections_idx = scipy.optimize.linear_sum_assignment(cost_matrix, maximize=True)
valid = cost_matrix[labels_idx, detections_idx] > 0
if valid.any():
correct[detections_idx[valid], i] = True
else:
matches = np.nonzero(iou >= threshold) # IoU > threshold and classes match
matches = np.array(matches).T
if matches.shape[0]:
if matches.shape[0] > 1:
matches = matches[iou[matches[:, 0], matches[:, 1]].argsort()[::-1]]
matches = matches[np.unique(matches[:, 1], return_index=True)[1]]
# matches = matches[matches[:, 2].argsort()[::-1]]
matches = matches[np.unique(matches[:, 0], return_index=True)[1]]
correct[matches[:, 1].astype(int), i] = True
return torch.tensor(correct, dtype=torch.bool, device=pred_classes.device)
def add_callback(self, event: str, callback):
"""Appends the given callback."""
self.callbacks[event].append(callback)
def run_callbacks(self, event: str):
"""Runs all callbacks associated with a specified event."""
for callback in self.callbacks.get(event, []):
callback(self)
def get_dataloader(self, dataset_path, batch_size):
"""Get data loader from dataset path and batch size."""
raise NotImplementedError("get_dataloader function not implemented for this validator")
def build_dataset(self, img_path):
"""Build dataset."""
raise NotImplementedError("build_dataset function not implemented in validator")
def preprocess(self, batch):
"""Preprocesses an input batch."""
return batch
def postprocess(self, preds):
"""Describes and summarizes the purpose of 'postprocess()' but no details mentioned."""
return preds
def init_metrics(self, model):
"""Initialize performance metrics for the YOLO model."""
pass
def update_metrics(self, preds, batch):
"""Updates metrics based on predictions and batch."""
pass
def finalize_metrics(self, *args, **kwargs):
"""Finalizes and returns all metrics."""
pass
def get_stats(self):
"""Returns statistics about the model's performance."""
return {}
def check_stats(self, stats):
"""Checks statistics."""
pass
def print_results(self):
"""Prints the results of the model's predictions."""
pass
def get_desc(self):
"""Get description of the YOLO model."""
pass
@property
def metric_keys(self):
"""Returns the metric keys used in YOLO training/validation."""
return []
def on_plot(self, name, data=None):
"""Registers plots (e.g. to be consumed in callbacks)"""
self.plots[Path(name)] = {"data": data, "timestamp": time.time()}
# TODO: may need to put these following functions into callback
def plot_val_samples(self, batch, ni):
"""Plots validation samples during training."""
pass
def plot_predictions(self, batch, preds, ni):
"""Plots YOLO model predictions on batch images."""
pass
def pred_to_json(self, preds, batch):
"""Convert predictions to JSON format."""
pass
def eval_json(self, stats):
"""Evaluate and return JSON format of prediction statistics."""
pass

128
ultralytics/hub/__init__.py
View File

@ -1,128 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import requests
from ultralytics.data.utils import HUBDatasetStats
from ultralytics.hub.auth import Auth
from ultralytics.hub.utils import HUB_API_ROOT, HUB_WEB_ROOT, PREFIX
from ultralytics.utils import LOGGER, SETTINGS, checks
def login(api_key: str = None, save=True) -> bool:
"""
Log in to the Ultralytics HUB API using the provided API key.
The session is not stored; a new session is created when needed using the saved SETTINGS or the HUB_API_KEY
environment variable if successfully authenticated.
Args:
api_key (str, optional): API key to use for authentication.
If not provided, it will be retrieved from SETTINGS or HUB_API_KEY environment variable.
save (bool, optional): Whether to save the API key to SETTINGS if authentication is successful.
Returns:
(bool): True if authentication is successful, False otherwise.
"""
checks.check_requirements("hub-sdk>=0.0.6")
from hub_sdk import HUBClient
api_key_url = f"{HUB_WEB_ROOT}/settings?tab=api+keys" # set the redirect URL
saved_key = SETTINGS.get("api_key")
active_key = api_key or saved_key
credentials = {"api_key": active_key} if active_key and active_key != "" else None # set credentials
client = HUBClient(credentials) # initialize HUBClient
if client.authenticated:
# Successfully authenticated with HUB
if save and client.api_key != saved_key:
SETTINGS.update({"api_key": client.api_key}) # update settings with valid API key
# Set message based on whether key was provided or retrieved from settings
log_message = (
"New authentication successful ✅" if client.api_key == api_key or not credentials else "Authenticated ✅"
)
LOGGER.info(f"{PREFIX}{log_message}")
return True
else:
# Failed to authenticate with HUB
LOGGER.info(f"{PREFIX}Get API key from {api_key_url} and then run 'yolo hub login API_KEY'")
return False
def logout():
"""
Log out of Ultralytics HUB by removing the API key from the settings file. To log in again, use 'yolo hub login'.
Example:
```python
from ultralytics import hub
hub.logout()
```
"""
SETTINGS["api_key"] = ""
SETTINGS.save()
LOGGER.info(f"{PREFIX}logged out ✅. To log in again, use 'yolo hub login'.")
def reset_model(model_id=""):
"""Reset a trained model to an untrained state."""
r = requests.post(f"{HUB_API_ROOT}/model-reset", json={"modelId": model_id}, headers={"x-api-key": Auth().api_key})
if r.status_code == 200:
LOGGER.info(f"{PREFIX}Model reset successfully")
return
LOGGER.warning(f"{PREFIX}Model reset failure {r.status_code} {r.reason}")
def export_fmts_hub():
"""Returns a list of HUB-supported export formats."""
from ultralytics.engine.exporter import export_formats
return list(export_formats()["Argument"][1:]) + ["ultralytics_tflite", "ultralytics_coreml"]
def export_model(model_id="", format="torchscript"):
"""Export a model to all formats."""
assert format in export_fmts_hub(), f"Unsupported export format '{format}', valid formats are {export_fmts_hub()}"
r = requests.post(
f"{HUB_API_ROOT}/v1/models/{model_id}/export", json={"format": format}, headers={"x-api-key": Auth().api_key}
)
assert r.status_code == 200, f"{PREFIX}{format} export failure {r.status_code} {r.reason}"
LOGGER.info(f"{PREFIX}{format} export started ✅")
def get_export(model_id="", format="torchscript"):
"""Get an exported model dictionary with download URL."""
assert format in export_fmts_hub(), f"Unsupported export format '{format}', valid formats are {export_fmts_hub()}"
r = requests.post(
f"{HUB_API_ROOT}/get-export",
json={"apiKey": Auth().api_key, "modelId": model_id, "format": format},
headers={"x-api-key": Auth().api_key},
)
assert r.status_code == 200, f"{PREFIX}{format} get_export failure {r.status_code} {r.reason}"
return r.json()
def check_dataset(path="", task="detect"):
"""
Function for error-checking HUB dataset Zip file before upload. It checks a dataset for errors before it is uploaded
to the HUB. Usage examples are given below.
Args:
path (str, optional): Path to data.zip (with data.yaml inside data.zip). Defaults to ''.
task (str, optional): Dataset task. Options are 'detect', 'segment', 'pose', 'classify'. Defaults to 'detect'.
Example:
```python
from ultralytics.hub import check_dataset
check_dataset('path/to/coco8.zip', task='detect') # detect dataset
check_dataset('path/to/coco8-seg.zip', task='segment') # segment dataset
check_dataset('path/to/coco8-pose.zip', task='pose') # pose dataset
```
"""
HUBDatasetStats(path=path, task=task).get_json()
LOGGER.info(f"Checks completed correctly ✅. Upload this dataset to {HUB_WEB_ROOT}/datasets/.")

136
ultralytics/hub/auth.py
View File

@ -1,136 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import requests
from ultralytics.hub.utils import HUB_API_ROOT, HUB_WEB_ROOT, PREFIX, request_with_credentials
from ultralytics.utils import LOGGER, SETTINGS, emojis, is_colab
API_KEY_URL = f"{HUB_WEB_ROOT}/settings?tab=api+keys"
class Auth:
"""
Manages authentication processes including API key handling, cookie-based authentication, and header generation.
The class supports different methods of authentication:
1. Directly using an API key.
2. Authenticating using browser cookies (specifically in Google Colab).
3. Prompting the user to enter an API key.
Attributes:
id_token (str or bool): Token used for identity verification, initialized as False.
api_key (str or bool): API key for authentication, initialized as False.
model_key (bool): Placeholder for model key, initialized as False.
"""
id_token = api_key = model_key = False
def __init__(self, api_key="", verbose=False):
"""
Initialize the Auth class with an optional API key.
Args:
api_key (str, optional): May be an API key or a combination API key and model ID, i.e. key_id
"""
# Split the input API key in case it contains a combined key_model and keep only the API key part
api_key = api_key.split("_")[0]
# Set API key attribute as value passed or SETTINGS API key if none passed
self.api_key = api_key or SETTINGS.get("api_key", "")
# If an API key is provided
if self.api_key:
# If the provided API key matches the API key in the SETTINGS
if self.api_key == SETTINGS.get("api_key"):
# Log that the user is already logged in
if verbose:
LOGGER.info(f"{PREFIX}Authenticated ✅")
return
else:
# Attempt to authenticate with the provided API key
success = self.authenticate()
# If the API key is not provided and the environment is a Google Colab notebook
elif is_colab():
# Attempt to authenticate using browser cookies
success = self.auth_with_cookies()
else:
# Request an API key
success = self.request_api_key()
# Update SETTINGS with the new API key after successful authentication
if success:
SETTINGS.update({"api_key": self.api_key})
# Log that the new login was successful
if verbose:
LOGGER.info(f"{PREFIX}New authentication successful ✅")
elif verbose:
LOGGER.info(f"{PREFIX}Get API key from {API_KEY_URL} and then run 'yolo hub login API_KEY'")
def request_api_key(self, max_attempts=3):
"""
Prompt the user to input their API key.
Returns the model ID.
"""
import getpass
for attempts in range(max_attempts):
LOGGER.info(f"{PREFIX}Login. Attempt {attempts + 1} of {max_attempts}")
input_key = getpass.getpass(f"Enter API key from {API_KEY_URL} ")
self.api_key = input_key.split("_")[0] # remove model id if present
if self.authenticate():
return True
raise ConnectionError(emojis(f"{PREFIX}Failed to authenticate ❌"))
def authenticate(self) -> bool:
"""
Attempt to authenticate with the server using either id_token or API key.
Returns:
(bool): True if authentication is successful, False otherwise.
"""
try:
if header := self.get_auth_header():
r = requests.post(f"{HUB_API_ROOT}/v1/auth", headers=header)
if not r.json().get("success", False):
raise ConnectionError("Unable to authenticate.")
return True
raise ConnectionError("User has not authenticated locally.")
except ConnectionError:
self.id_token = self.api_key = False # reset invalid
LOGGER.warning(f"{PREFIX}Invalid API key ⚠️")
return False
def auth_with_cookies(self) -> bool:
"""
Attempt to fetch authentication via cookies and set id_token. User must be logged in to HUB and running in a
supported browser.
Returns:
(bool): True if authentication is successful, False otherwise.
"""
if not is_colab():
return False # Currently only works with Colab
try:
authn = request_with_credentials(f"{HUB_API_ROOT}/v1/auth/auto")
if authn.get("success", False):
self.id_token = authn.get("data", {}).get("idToken", None)
self.authenticate()
return True
raise ConnectionError("Unable to fetch browser authentication details.")
except ConnectionError:
self.id_token = False # reset invalid
return False
def get_auth_header(self):
"""
Get the authentication header for making API requests.
Returns:
(dict): The authentication header if id_token or API key is set, None otherwise.
"""
if self.id_token:
return {"authorization": f"Bearer {self.id_token}"}
elif self.api_key:
return {"x-api-key": self.api_key}
# else returns None

355
ultralytics/hub/session.py
View File

@ -1,355 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import threading
import time
from http import HTTPStatus
from pathlib import Path
import requests
from ultralytics.hub.utils import HUB_WEB_ROOT, HELP_MSG, PREFIX, TQDM
from ultralytics.utils import LOGGER, SETTINGS, __version__, checks, emojis, is_colab
from ultralytics.utils.errors import HUBModelError
AGENT_NAME = f"python-{__version__}-colab" if is_colab() else f"python-{__version__}-local"
class HUBTrainingSession:
"""
HUB training session for Ultralytics HUB YOLO models. Handles model initialization, heartbeats, and checkpointing.
Attributes:
agent_id (str): Identifier for the instance communicating with the server.
model_id (str): Identifier for the YOLO model being trained.
model_url (str): URL for the model in Ultralytics HUB.
api_url (str): API URL for the model in Ultralytics HUB.
auth_header (dict): Authentication header for the Ultralytics HUB API requests.
rate_limits (dict): Rate limits for different API calls (in seconds).
timers (dict): Timers for rate limiting.
metrics_queue (dict): Queue for the model's metrics.
model (dict): Model data fetched from Ultralytics HUB.
alive (bool): Indicates if the heartbeat loop is active.
"""
def __init__(self, identifier):
"""
Initialize the HUBTrainingSession with the provided model identifier.
Args:
identifier (str): Model identifier used to initialize the HUB training session.
It can be a URL string or a model key with specific format.
Raises:
ValueError: If the provided model identifier is invalid.
ConnectionError: If connecting with global API key is not supported.
ModuleNotFoundError: If hub-sdk package is not installed.
"""
from hub_sdk import HUBClient
self.rate_limits = {
"metrics": 3.0,
"ckpt": 900.0,
"heartbeat": 300.0,
} # rate limits (seconds)
self.metrics_queue = {} # holds metrics for each epoch until upload
self.metrics_upload_failed_queue = {} # holds metrics for each epoch if upload failed
self.timers = {} # holds timers in ultralytics/utils/callbacks/hub.py
# Parse input
api_key, model_id, self.filename = self._parse_identifier(identifier)
# Get credentials
active_key = api_key or SETTINGS.get("api_key")
credentials = {"api_key": active_key} if active_key else None # set credentials
# Initialize client
self.client = HUBClient(credentials)
if model_id:
self.load_model(model_id) # load existing model
else:
self.model = self.client.model() # load empty model
def load_model(self, model_id):
"""Loads an existing model from Ultralytics HUB using the provided model identifier."""
self.model = self.client.model(model_id)
if not self.model.data: # then model does not exist
raise ValueError(emojis("❌ The specified HUB model does not exist")) # TODO: improve error handling
self.model_url = f"{HUB_WEB_ROOT}/models/{self.model.id}"
self._set_train_args()
# Start heartbeats for HUB to monitor agent
self.model.start_heartbeat(self.rate_limits["heartbeat"])
LOGGER.info(f"{PREFIX}View model at {self.model_url} 🚀")
def create_model(self, model_args):
"""Initializes a HUB training session with the specified model identifier."""
payload = {
"config": {
"batchSize": model_args.get("batch", -1),
"epochs": model_args.get("epochs", 300),
"imageSize": model_args.get("imgsz", 640),
"patience": model_args.get("patience", 100),
"device": model_args.get("device", ""),
"cache": model_args.get("cache", "ram"),
},
"dataset": {"name": model_args.get("data")},
"lineage": {
"architecture": {
"name": self.filename.replace(".pt", "").replace(".yaml", ""),
},
"parent": {},
},
"meta": {"name": self.filename},
}
if self.filename.endswith(".pt"):
payload["lineage"]["parent"]["name"] = self.filename
self.model.create_model(payload)
# Model could not be created
# TODO: improve error handling
if not self.model.id:
return
self.model_url = f"{HUB_WEB_ROOT}/models/{self.model.id}"
# Start heartbeats for HUB to monitor agent
self.model.start_heartbeat(self.rate_limits["heartbeat"])
LOGGER.info(f"{PREFIX}View model at {self.model_url} 🚀")
def _parse_identifier(self, identifier):
"""
Parses the given identifier to determine the type of identifier and extract relevant components.
The method supports different identifier formats:
- A HUB URL, which starts with HUB_WEB_ROOT followed by '/models/'
- An identifier containing an API key and a model ID separated by an underscore
- An identifier that is solely a model ID of a fixed length
- A local filename that ends with '.pt' or '.yaml'
Args:
identifier (str): The identifier string to be parsed.
Returns:
(tuple): A tuple containing the API key, model ID, and filename as applicable.
Raises:
HUBModelError: If the identifier format is not recognized.
"""
# Initialize variables
api_key, model_id, filename = None, None, None
# Check if identifier is a HUB URL
if identifier.startswith(f"{HUB_WEB_ROOT}/models/"):
# Extract the model_id after the HUB_WEB_ROOT URL
model_id = identifier.split(f"{HUB_WEB_ROOT}/models/")[-1]
else:
# Split the identifier based on underscores only if it's not a HUB URL
parts = identifier.split("_")
# Check if identifier is in the format of API key and model ID
if len(parts) == 2 and len(parts[0]) == 42 and len(parts[1]) == 20:
api_key, model_id = parts
# Check if identifier is a single model ID
elif len(parts) == 1 and len(parts[0]) == 20:
model_id = parts[0]
# Check if identifier is a local filename
elif identifier.endswith(".pt") or identifier.endswith(".yaml"):
filename = identifier
else:
raise HUBModelError(
f"model='{identifier}' could not be parsed. Check format is correct. "
f"Supported formats are Ultralytics HUB URL, apiKey_modelId, modelId, local pt or yaml file."
)
return api_key, model_id, filename
def _set_train_args(self):
"""
Initializes training arguments and creates a model entry on the Ultralytics HUB.
This method sets up training arguments based on the model's state and updates them with any additional
arguments provided. It handles different states of the model, such as whether it's resumable, pretrained,
or requires specific file setup.
Raises:
ValueError: If the model is already trained, if required dataset information is missing, or if there are
issues with the provided training arguments.
"""
if self.model.is_trained():
raise ValueError(emojis(f"Model is already trained and uploaded to {self.model_url} 🚀"))
if self.model.is_resumable():
# Model has saved weights
self.train_args = {"data": self.model.get_dataset_url(), "resume": True}
self.model_file = self.model.get_weights_url("last")
else:
# Model has no saved weights
self.train_args = self.model.data.get("train_args") # new response
# Set the model file as either a *.pt or *.yaml file
self.model_file = (
self.model.get_weights_url("parent") if self.model.is_pretrained() else self.model.get_architecture()
)
if "data" not in self.train_args:
# RF bug - datasets are sometimes not exported
raise ValueError("Dataset may still be processing. Please wait a minute and try again.")
self.model_file = checks.check_yolov5u_filename(self.model_file, verbose=False) # YOLOv5->YOLOv5u
self.model_id = self.model.id
def request_queue(
self,
request_func,
retry=3,
timeout=30,
thread=True,
verbose=True,
progress_total=None,
*args,
**kwargs,
):
def retry_request():
"""Attempts to call `request_func` with retries, timeout, and optional threading."""
t0 = time.time() # Record the start time for the timeout
for i in range(retry + 1):
if (time.time() - t0) > timeout:
LOGGER.warning(f"{PREFIX}Timeout for request reached. {HELP_MSG}")
break # Timeout reached, exit loop
response = request_func(*args, **kwargs)
if response is None:
LOGGER.warning(f"{PREFIX}Received no response from the request. {HELP_MSG}")
time.sleep(2**i) # Exponential backoff before retrying
continue # Skip further processing and retry
if progress_total:
self._show_upload_progress(progress_total, response)
if HTTPStatus.OK <= response.status_code < HTTPStatus.MULTIPLE_CHOICES:
# if request related to metrics upload
if kwargs.get("metrics"):
self.metrics_upload_failed_queue = {}
return response # Success, no need to retry
if i == 0:
# Initial attempt, check status code and provide messages
message = self._get_failure_message(response, retry, timeout)
if verbose:
LOGGER.warning(f"{PREFIX}{message} {HELP_MSG} ({response.status_code})")
if not self._should_retry(response.status_code):
LOGGER.warning(f"{PREFIX}Request failed. {HELP_MSG} ({response.status_code}")
break # Not an error that should be retried, exit loop
time.sleep(2**i) # Exponential backoff for retries
# if request related to metrics upload and exceed retries
if response is None and kwargs.get("metrics"):
self.metrics_upload_failed_queue.update(kwargs.get("metrics", None))
return response
if thread:
# Start a new thread to run the retry_request function
threading.Thread(target=retry_request, daemon=True).start()
else:
# If running in the main thread, call retry_request directly
return retry_request()
def _should_retry(self, status_code):
"""Determines if a request should be retried based on the HTTP status code."""
retry_codes = {
HTTPStatus.REQUEST_TIMEOUT,
HTTPStatus.BAD_GATEWAY,
HTTPStatus.GATEWAY_TIMEOUT,
}
return status_code in retry_codes
def _get_failure_message(self, response: requests.Response, retry: int, timeout: int):
"""
Generate a retry message based on the response status code.
Args:
response: The HTTP response object.
retry: The number of retry attempts allowed.
timeout: The maximum timeout duration.
Returns:
(str): The retry message.
"""
if self._should_retry(response.status_code):
return f"Retrying {retry}x for {timeout}s." if retry else ""
elif response.status_code == HTTPStatus.TOO_MANY_REQUESTS: # rate limit
headers = response.headers
return (
f"Rate limit reached ({headers['X-RateLimit-Remaining']}/{headers['X-RateLimit-Limit']}). "
f"Please retry after {headers['Retry-After']}s."
)
else:
try:
return response.json().get("message", "No JSON message.")
except AttributeError:
return "Unable to read JSON."
def upload_metrics(self):
"""Upload model metrics to Ultralytics HUB."""
return self.request_queue(self.model.upload_metrics, metrics=self.metrics_queue.copy(), thread=True)
def upload_model(
self,
epoch: int,
weights: str,
is_best: bool = False,
map: float = 0.0,
final: bool = False,
) -> None:
"""
Upload a model checkpoint to Ultralytics HUB.
Args:
epoch (int): The current training epoch.
weights (str): Path to the model weights file.
is_best (bool): Indicates if the current model is the best one so far.
map (float): Mean average precision of the model.
final (bool): Indicates if the model is the final model after training.
"""
if Path(weights).is_file():
progress_total = Path(weights).stat().st_size if final else None # Only show progress if final
self.request_queue(
self.model.upload_model,
epoch=epoch,
weights=weights,
is_best=is_best,
map=map,
final=final,
retry=10,
timeout=3600,
thread=not final,
progress_total=progress_total,
)
else:
LOGGER.warning(f"{PREFIX}WARNING ⚠️ Model upload issue. Missing model {weights}.")
def _show_upload_progress(self, content_length: int, response: requests.Response) -> None:
"""
Display a progress bar to track the upload progress of a file download.
Args:
content_length (int): The total size of the content to be downloaded in bytes.
response (requests.Response): The response object from the file download request.
Returns:
None
"""
with TQDM(total=content_length, unit="B", unit_scale=True, unit_divisor=1024) as pbar:
for data in response.iter_content(chunk_size=1024):
pbar.update(len(data))

247
ultralytics/hub/utils.py
View File

@ -1,247 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import os
import platform
import random
import sys
import threading
import time
from pathlib import Path
import requests
from ultralytics.utils import (
ENVIRONMENT,
LOGGER,
ONLINE,
RANK,
SETTINGS,
TESTS_RUNNING,
TQDM,
TryExcept,
__version__,
colorstr,
get_git_origin_url,
is_colab,
is_git_dir,
is_pip_package,
)
from ultralytics.utils.downloads import GITHUB_ASSETS_NAMES
HUB_API_ROOT = os.environ.get("ULTRALYTICS_HUB_API", "https://api.ultralytics.com")
HUB_WEB_ROOT = os.environ.get("ULTRALYTICS_HUB_WEB", "https://hub.ultralytics.com")
PREFIX = colorstr("Ultralytics HUB: ")
HELP_MSG = "If this issue persists please visit https://github.com/ultralytics/hub/issues for assistance."
def request_with_credentials(url: str) -> any:
"""
Make an AJAX request with cookies attached in a Google Colab environment.
Args:
url (str): The URL to make the request to.
Returns:
(any): The response data from the AJAX request.
Raises:
OSError: If the function is not run in a Google Colab environment.
"""
if not is_colab():
raise OSError("request_with_credentials() must run in a Colab environment")
from google.colab import output # noqa
from IPython import display # noqa
display.display(
display.Javascript(
"""
window._hub_tmp = new Promise((resolve, reject) => {
const timeout = setTimeout(() => reject("Failed authenticating existing browser session"), 5000)
fetch("%s", {
method: 'POST',
credentials: 'include'
})
.then((response) => resolve(response.json()))
.then((json) => {
clearTimeout(timeout);
}).catch((err) => {
clearTimeout(timeout);
reject(err);
});
});
"""
% url
)
)
return output.eval_js("_hub_tmp")
def requests_with_progress(method, url, **kwargs):
"""
Make an HTTP request using the specified method and URL, with an optional progress bar.
Args:
method (str): The HTTP method to use (e.g. 'GET', 'POST').
url (str): The URL to send the request to.
**kwargs (any): Additional keyword arguments to pass to the underlying `requests.request` function.
Returns:
(requests.Response): The response object from the HTTP request.
Note:
- If 'progress' is set to True, the progress bar will display the download progress for responses with a known
content length.
- If 'progress' is a number then progress bar will display assuming content length = progress.
"""
progress = kwargs.pop("progress", False)
if not progress:
return requests.request(method, url, **kwargs)
response = requests.request(method, url, stream=True, **kwargs)
total = int(response.headers.get("content-length", 0) if isinstance(progress, bool) else progress) # total size
try:
pbar = TQDM(total=total, unit="B", unit_scale=True, unit_divisor=1024)
for data in response.iter_content(chunk_size=1024):
pbar.update(len(data))
pbar.close()
except requests.exceptions.ChunkedEncodingError: # avoid 'Connection broken: IncompleteRead' warnings
response.close()
return response
def smart_request(method, url, retry=3, timeout=30, thread=True, code=-1, verbose=True, progress=False, **kwargs):
"""
Makes an HTTP request using the 'requests' library, with exponential backoff retries up to a specified timeout.
Args:
method (str): The HTTP method to use for the request. Choices are 'post' and 'get'.
url (str): The URL to make the request to.
retry (int, optional): Number of retries to attempt before giving up. Default is 3.
timeout (int, optional): Timeout in seconds after which the function will give up retrying. Default is 30.
thread (bool, optional): Whether to execute the request in a separate daemon thread. Default is True.
code (int, optional): An identifier for the request, used for logging purposes. Default is -1.
verbose (bool, optional): A flag to determine whether to print out to console or not. Default is True.
progress (bool, optional): Whether to show a progress bar during the request. Default is False.
**kwargs (any): Keyword arguments to be passed to the requests function specified in method.
Returns:
(requests.Response): The HTTP response object. If the request is executed in a separate thread, returns None.
"""
retry_codes = (408, 500) # retry only these codes
@TryExcept(verbose=verbose)
def func(func_method, func_url, **func_kwargs):
"""Make HTTP requests with retries and timeouts, with optional progress tracking."""
r = None # response
t0 = time.time() # initial time for timer
for i in range(retry + 1):
if (time.time() - t0) > timeout:
break
r = requests_with_progress(func_method, func_url, **func_kwargs) # i.e. get(url, data, json, files)
if r.status_code < 300: # return codes in the 2xx range are generally considered "good" or "successful"
break
try:
m = r.json().get("message", "No JSON message.")
except AttributeError:
m = "Unable to read JSON."
if i == 0:
if r.status_code in retry_codes:
m += f" Retrying {retry}x for {timeout}s." if retry else ""
elif r.status_code == 429: # rate limit
h = r.headers # response headers
m = (
f"Rate limit reached ({h['X-RateLimit-Remaining']}/{h['X-RateLimit-Limit']}). "
f"Please retry after {h['Retry-After']}s."
)
if verbose:
LOGGER.warning(f"{PREFIX}{m} {HELP_MSG} ({r.status_code} #{code})")
if r.status_code not in retry_codes:
return r
time.sleep(2**i) # exponential standoff
return r
args = method, url
kwargs["progress"] = progress
if thread:
threading.Thread(target=func, args=args, kwargs=kwargs, daemon=True).start()
else:
return func(*args, **kwargs)
class Events:
"""
A class for collecting anonymous event analytics. Event analytics are enabled when sync=True in settings and
disabled when sync=False. Run 'yolo settings' to see and update settings YAML file.
Attributes:
url (str): The URL to send anonymous events.
rate_limit (float): The rate limit in seconds for sending events.
metadata (dict): A dictionary containing metadata about the environment.
enabled (bool): A flag to enable or disable Events based on certain conditions.
"""
url = "https://www.google-analytics.com/mp/collect?measurement_id=G-X8NCJYTQXM&api_secret=QLQrATrNSwGRFRLE-cbHJw"
def __init__(self):
"""Initializes the Events object with default values for events, rate_limit, and metadata."""
self.events = [] # events list
self.rate_limit = 60.0 # rate limit (seconds)
self.t = 0.0 # rate limit timer (seconds)
self.metadata = {
"cli": Path(sys.argv[0]).name == "yolo",
"install": "git" if is_git_dir() else "pip" if is_pip_package() else "other",
"python": ".".join(platform.python_version_tuple()[:2]), # i.e. 3.10
"version": __version__,
"env": ENVIRONMENT,
"session_id": round(random.random() * 1e15),
"engagement_time_msec": 1000,
}
self.enabled = (
SETTINGS["sync"]
and RANK in (-1, 0)
and not TESTS_RUNNING
and ONLINE
and (is_pip_package() or get_git_origin_url() == "https://github.com/ultralytics/ultralytics.git")
)
def __call__(self, cfg):
"""
Attempts to add a new event to the events list and send events if the rate limit is reached.
Args:
cfg (IterableSimpleNamespace): The configuration object containing mode and task information.
"""
if not self.enabled:
# Events disabled, do nothing
return
# Attempt to add to events
if len(self.events) < 25: # Events list limited to 25 events (drop any events past this)
params = {
**self.metadata,
"task": cfg.task,
"model": cfg.model if cfg.model in GITHUB_ASSETS_NAMES else "custom",
}
if cfg.mode == "export":
params["format"] = cfg.format
self.events.append({"name": cfg.mode, "params": params})
# Check rate limit
t = time.time()
if (t - self.t) < self.rate_limit:
# Time is under rate limiter, wait to send
return
# Time is over rate limiter, send now
data = {"client_id": SETTINGS["uuid"], "events": self.events} # SHA-256 anonymized UUID hash and events list
# POST equivalent to requests.post(self.url, json=data)
smart_request("post", self.url, json=data, retry=0, verbose=False)
# Reset events and rate limit timer
self.events = []
self.t = t
# Run below code on hub/utils init -------------------------------------------------------------------------------------
events = Events()

8
ultralytics/models/__init__.py
View File

@ -1,8 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
from .rtdetr import RTDETR
from .sam import SAM
from .yolo import YOLO, YOLOWorld
from .yolov10 import YOLOv10
from .yolov10seg import YOLOv10Seg
__all__ = "YOLO", "RTDETR", "SAM", "YOLOWorld", "YOLOv10","YOLOv10Seg" # allow simpler import

8
ultralytics/models/fastsam/__init__.py
View File

@ -1,8 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
from .model import FastSAM
from .predict import FastSAMPredictor
from .prompt import FastSAMPrompt
from .val import FastSAMValidator
__all__ = "FastSAMPredictor", "FastSAM", "FastSAMPrompt", "FastSAMValidator"

33
ultralytics/models/fastsam/model.py
View File

@ -1,33 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
from pathlib import Path
from ultralytics.engine.model import Model
from .predict import FastSAMPredictor
from .val import FastSAMValidator
class FastSAM(Model):
"""
FastSAM model interface.
Example:
```python
from ultralytics import FastSAM
model = FastSAM('last.pt')
results = model.predict('ultralytics/assets/bus.jpg')
```
"""
def __init__(self, model="FastSAM-x.pt"):
"""Call the __init__ method of the parent class (YOLO) with the updated default model."""
if str(model) == "FastSAM.pt":
model = "FastSAM-x.pt"
assert Path(model).suffix not in (".yaml", ".yml"), "FastSAM models only support pre-trained models."
super().__init__(model=model, task="segment")
@property
def task_map(self):
"""Returns a dictionary mapping segment task to corresponding predictor and validator classes."""
return {"segment": {"predictor": FastSAMPredictor, "validator": FastSAMValidator}}

86
ultralytics/models/fastsam/predict.py
View File

@ -1,86 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import torch
from ultralytics.engine.results import Results
from ultralytics.models.fastsam.utils import bbox_iou
from ultralytics.models.yolo.detect.predict import DetectionPredictor
from ultralytics.utils import DEFAULT_CFG, ops
class FastSAMPredictor(DetectionPredictor):
"""
FastSAMPredictor is specialized for fast SAM (Segment Anything Model) segmentation prediction tasks in Ultralytics
YOLO framework.
This class extends the DetectionPredictor, customizing the prediction pipeline specifically for fast SAM.
It adjusts post-processing steps to incorporate mask prediction and non-max suppression while optimizing
for single-class segmentation.
Attributes:
cfg (dict): Configuration parameters for prediction.
overrides (dict, optional): Optional parameter overrides for custom behavior.
_callbacks (dict, optional): Optional list of callback functions to be invoked during prediction.
"""
def __init__(self, cfg=DEFAULT_CFG, overrides=None, _callbacks=None):
"""
Initializes the FastSAMPredictor class, inheriting from DetectionPredictor and setting the task to 'segment'.
Args:
cfg (dict): Configuration parameters for prediction.
overrides (dict, optional): Optional parameter overrides for custom behavior.
_callbacks (dict, optional): Optional list of callback functions to be invoked during prediction.
"""
super().__init__(cfg, overrides, _callbacks)
self.args.task = "segment"
def postprocess(self, preds, img, orig_imgs):
"""
Perform post-processing steps on predictions, including non-max suppression and scaling boxes to original image
size, and returns the final results.
Args:
preds (list): The raw output predictions from the model.
img (torch.Tensor): The processed image tensor.
orig_imgs (list | torch.Tensor): The original image or list of images.
Returns:
(list): A list of Results objects, each containing processed boxes, masks, and other metadata.
"""
p = ops.non_max_suppression(
preds[0],
self.args.conf,
self.args.iou,
agnostic=self.args.agnostic_nms,
max_det=self.args.max_det,
nc=1, # set to 1 class since SAM has no class predictions
classes=self.args.classes,
)
full_box = torch.zeros(p[0].shape[1], device=p[0].device)
full_box[2], full_box[3], full_box[4], full_box[6:] = img.shape[3], img.shape[2], 1.0, 1.0
full_box = full_box.view(1, -1)
critical_iou_index = bbox_iou(full_box[0][:4], p[0][:, :4], iou_thres=0.9, image_shape=img.shape[2:])
if critical_iou_index.numel() != 0:
full_box[0][4] = p[0][critical_iou_index][:, 4]
full_box[0][6:] = p[0][critical_iou_index][:, 6:]
p[0][critical_iou_index] = full_box
if not isinstance(orig_imgs, list): # input images are a torch.Tensor, not a list
orig_imgs = ops.convert_torch2numpy_batch(orig_imgs)
results = []
proto = preds[1][-1] if len(preds[1]) == 3 else preds[1] # second output is len 3 if pt, but only 1 if exported
for i, pred in enumerate(p):
orig_img = orig_imgs[i]
img_path = self.batch[0][i]
if not len(pred): # save empty boxes
masks = None
elif self.args.retina_masks:
pred[:, :4] = ops.scale_boxes(img.shape[2:], pred[:, :4], orig_img.shape)
masks = ops.process_mask_native(proto[i], pred[:, 6:], pred[:, :4], orig_img.shape[:2]) # HWC
else:
masks = ops.process_mask(proto[i], pred[:, 6:], pred[:, :4], img.shape[2:], upsample=True) # HWC
pred[:, :4] = ops.scale_boxes(img.shape[2:], pred[:, :4], orig_img.shape)
results.append(Results(orig_img, path=img_path, names=self.model.names, boxes=pred[:, :6], masks=masks))
return results

357
ultralytics/models/fastsam/prompt.py
View File

@ -1,357 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import os
from pathlib import Path
import cv2
import matplotlib.pyplot as plt
import numpy as np
import torch
from PIL import Image
from ultralytics.utils import TQDM
class FastSAMPrompt:
"""
Fast Segment Anything Model class for image annotation and visualization.
Attributes:
device (str): Computing device ('cuda' or 'cpu').
results: Object detection or segmentation results.
source: Source image or image path.
clip: CLIP model for linear assignment.
"""
def __init__(self, source, results, device="cuda") -> None:
"""Initializes FastSAMPrompt with given source, results and device, and assigns clip for linear assignment."""
self.device = device
self.results = results
self.source = source
# Import and assign clip
try:
import clip
except ImportError:
from ultralytics.utils.checks import check_requirements
check_requirements("git+https://github.com/openai/CLIP.git")
import clip
self.clip = clip
@staticmethod
def _segment_image(image, bbox):
"""Segments the given image according to the provided bounding box coordinates."""
image_array = np.array(image)
segmented_image_array = np.zeros_like(image_array)
x1, y1, x2, y2 = bbox
segmented_image_array[y1:y2, x1:x2] = image_array[y1:y2, x1:x2]
segmented_image = Image.fromarray(segmented_image_array)
black_image = Image.new("RGB", image.size, (255, 255, 255))
# transparency_mask = np.zeros_like((), dtype=np.uint8)
transparency_mask = np.zeros((image_array.shape[0], image_array.shape[1]), dtype=np.uint8)
transparency_mask[y1:y2, x1:x2] = 255
transparency_mask_image = Image.fromarray(transparency_mask, mode="L")
black_image.paste(segmented_image, mask=transparency_mask_image)
return black_image
@staticmethod
def _format_results(result, filter=0):
"""Formats detection results into list of annotations each containing ID, segmentation, bounding box, score and
area.
"""
annotations = []
n = len(result.masks.data) if result.masks is not None else 0
for i in range(n):
mask = result.masks.data[i] == 1.0
if torch.sum(mask) >= filter:
annotation = {
"id": i,
"segmentation": mask.cpu().numpy(),
"bbox": result.boxes.data[i],
"score": result.boxes.conf[i],
}
annotation["area"] = annotation["segmentation"].sum()
annotations.append(annotation)
return annotations
@staticmethod
def _get_bbox_from_mask(mask):
"""Applies morphological transformations to the mask, displays it, and if with_contours is True, draws
contours.
"""
mask = mask.astype(np.uint8)
contours, hierarchy = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
x1, y1, w, h = cv2.boundingRect(contours[0])
x2, y2 = x1 + w, y1 + h
if len(contours) > 1:
for b in contours:
x_t, y_t, w_t, h_t = cv2.boundingRect(b)
x1 = min(x1, x_t)
y1 = min(y1, y_t)
x2 = max(x2, x_t + w_t)
y2 = max(y2, y_t + h_t)
return [x1, y1, x2, y2]
def plot(
self,
annotations,
output,
bbox=None,
points=None,
point_label=None,
mask_random_color=True,
better_quality=True,
retina=False,
with_contours=True,
):
"""
Plots annotations, bounding boxes, and points on images and saves the output.
Args:
annotations (list): Annotations to be plotted.
output (str or Path): Output directory for saving the plots.
bbox (list, optional): Bounding box coordinates [x1, y1, x2, y2]. Defaults to None.
points (list, optional): Points to be plotted. Defaults to None.
point_label (list, optional): Labels for the points. Defaults to None.
mask_random_color (bool, optional): Whether to use random color for masks. Defaults to True.
better_quality (bool, optional): Whether to apply morphological transformations for better mask quality. Defaults to True.
retina (bool, optional): Whether to use retina mask. Defaults to False.
with_contours (bool, optional): Whether to plot contours. Defaults to True.
"""
pbar = TQDM(annotations, total=len(annotations))
for ann in pbar:
result_name = os.path.basename(ann.path)
image = ann.orig_img[..., ::-1] # BGR to RGB
original_h, original_w = ann.orig_shape
# For macOS only
# plt.switch_backend('TkAgg')
plt.figure(figsize=(original_w / 100, original_h / 100))
# Add subplot with no margin.
plt.subplots_adjust(top=1, bottom=0, right=1, left=0, hspace=0, wspace=0)
plt.margins(0, 0)
plt.gca().xaxis.set_major_locator(plt.NullLocator())
plt.gca().yaxis.set_major_locator(plt.NullLocator())
plt.imshow(image)
if ann.masks is not None:
masks = ann.masks.data
if better_quality:
if isinstance(masks[0], torch.Tensor):
masks = np.array(masks.cpu())
for i, mask in enumerate(masks):
mask = cv2.morphologyEx(mask.astype(np.uint8), cv2.MORPH_CLOSE, np.ones((3, 3), np.uint8))
masks[i] = cv2.morphologyEx(mask.astype(np.uint8), cv2.MORPH_OPEN, np.ones((8, 8), np.uint8))
self.fast_show_mask(
masks,
plt.gca(),
random_color=mask_random_color,
bbox=bbox,
points=points,
pointlabel=point_label,
retinamask=retina,
target_height=original_h,
target_width=original_w,
)
if with_contours:
contour_all = []
temp = np.zeros((original_h, original_w, 1))
for i, mask in enumerate(masks):
mask = mask.astype(np.uint8)
if not retina:
mask = cv2.resize(mask, (original_w, original_h), interpolation=cv2.INTER_NEAREST)
contours, _ = cv2.findContours(mask, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
contour_all.extend(iter(contours))
cv2.drawContours(temp, contour_all, -1, (255, 255, 255), 2)
color = np.array([0 / 255, 0 / 255, 1.0, 0.8])
contour_mask = temp / 255 * color.reshape(1, 1, -1)
plt.imshow(contour_mask)
# Save the figure
save_path = Path(output) / result_name
save_path.parent.mkdir(exist_ok=True, parents=True)
plt.axis("off")
plt.savefig(save_path, bbox_inches="tight", pad_inches=0, transparent=True)
plt.close()
pbar.set_description(f"Saving {result_name} to {save_path}")
@staticmethod
def fast_show_mask(
annotation,
ax,
random_color=False,
bbox=None,
points=None,
pointlabel=None,
retinamask=True,
target_height=960,
target_width=960,
):
"""
Quickly shows the mask annotations on the given matplotlib axis.
Args:
annotation (array-like): Mask annotation.
ax (matplotlib.axes.Axes): Matplotlib axis.
random_color (bool, optional): Whether to use random color for masks. Defaults to False.
bbox (list, optional): Bounding box coordinates [x1, y1, x2, y2]. Defaults to None.
points (list, optional): Points to be plotted. Defaults to None.
pointlabel (list, optional): Labels for the points. Defaults to None.
retinamask (bool, optional): Whether to use retina mask. Defaults to True.
target_height (int, optional): Target height for resizing. Defaults to 960.
target_width (int, optional): Target width for resizing. Defaults to 960.
"""
n, h, w = annotation.shape # batch, height, width
areas = np.sum(annotation, axis=(1, 2))
annotation = annotation[np.argsort(areas)]
index = (annotation != 0).argmax(axis=0)
if random_color:
color = np.random.random((n, 1, 1, 3))
else:
color = np.ones((n, 1, 1, 3)) * np.array([30 / 255, 144 / 255, 1.0])
transparency = np.ones((n, 1, 1, 1)) * 0.6
visual = np.concatenate([color, transparency], axis=-1)
mask_image = np.expand_dims(annotation, -1) * visual
show = np.zeros((h, w, 4))
h_indices, w_indices = np.meshgrid(np.arange(h), np.arange(w), indexing="ij")
indices = (index[h_indices, w_indices], h_indices, w_indices, slice(None))
show[h_indices, w_indices, :] = mask_image[indices]
if bbox is not None:
x1, y1, x2, y2 = bbox
ax.add_patch(plt.Rectangle((x1, y1), x2 - x1, y2 - y1, fill=False, edgecolor="b", linewidth=1))
# Draw point
if points is not None:
plt.scatter(
[point[0] for i, point in enumerate(points) if pointlabel[i] == 1],
[point[1] for i, point in enumerate(points) if pointlabel[i] == 1],
s=20,
c="y",
)
plt.scatter(
[point[0] for i, point in enumerate(points) if pointlabel[i] == 0],
[point[1] for i, point in enumerate(points) if pointlabel[i] == 0],
s=20,
c="m",
)
if not retinamask:
show = cv2.resize(show, (target_width, target_height), interpolation=cv2.INTER_NEAREST)
ax.imshow(show)
@torch.no_grad()
def retrieve(self, model, preprocess, elements, search_text: str, device) -> int:
"""Processes images and text with a model, calculates similarity, and returns softmax score."""
preprocessed_images = [preprocess(image).to(device) for image in elements]
tokenized_text = self.clip.tokenize([search_text]).to(device)
stacked_images = torch.stack(preprocessed_images)
image_features = model.encode_image(stacked_images)
text_features = model.encode_text(tokenized_text)
image_features /= image_features.norm(dim=-1, keepdim=True)
text_features /= text_features.norm(dim=-1, keepdim=True)
probs = 100.0 * image_features @ text_features.T
return probs[:, 0].softmax(dim=0)
def _crop_image(self, format_results):
"""Crops an image based on provided annotation format and returns cropped images and related data."""
if os.path.isdir(self.source):
raise ValueError(f"'{self.source}' is a directory, not a valid source for this function.")
image = Image.fromarray(cv2.cvtColor(self.results[0].orig_img, cv2.COLOR_BGR2RGB))
ori_w, ori_h = image.size
annotations = format_results
mask_h, mask_w = annotations[0]["segmentation"].shape
if ori_w != mask_w or ori_h != mask_h:
image = image.resize((mask_w, mask_h))
cropped_boxes = []
cropped_images = []
not_crop = []
filter_id = []
for _, mask in enumerate(annotations):
if np.sum(mask["segmentation"]) <= 100:
filter_id.append(_)
continue
bbox = self._get_bbox_from_mask(mask["segmentation"]) # bbox from mask
cropped_boxes.append(self._segment_image(image, bbox)) # save cropped image
cropped_images.append(bbox) # save cropped image bbox
return cropped_boxes, cropped_images, not_crop, filter_id, annotations
def box_prompt(self, bbox):
"""Modifies the bounding box properties and calculates IoU between masks and bounding box."""
if self.results[0].masks is not None:
assert bbox[2] != 0 and bbox[3] != 0
if os.path.isdir(self.source):
raise ValueError(f"'{self.source}' is a directory, not a valid source for this function.")
masks = self.results[0].masks.data
target_height, target_width = self.results[0].orig_shape
h = masks.shape[1]
w = masks.shape[2]
if h != target_height or w != target_width:
bbox = [
int(bbox[0] * w / target_width),
int(bbox[1] * h / target_height),
int(bbox[2] * w / target_width),
int(bbox[3] * h / target_height),
]
bbox[0] = max(round(bbox[0]), 0)
bbox[1] = max(round(bbox[1]), 0)
bbox[2] = min(round(bbox[2]), w)
bbox[3] = min(round(bbox[3]), h)
# IoUs = torch.zeros(len(masks), dtype=torch.float32)
bbox_area = (bbox[3] - bbox[1]) * (bbox[2] - bbox[0])
masks_area = torch.sum(masks[:, bbox[1] : bbox[3], bbox[0] : bbox[2]], dim=(1, 2))
orig_masks_area = torch.sum(masks, dim=(1, 2))
union = bbox_area + orig_masks_area - masks_area
iou = masks_area / union
max_iou_index = torch.argmax(iou)
self.results[0].masks.data = torch.tensor(np.array([masks[max_iou_index].cpu().numpy()]))
return self.results
def point_prompt(self, points, pointlabel): # numpy
"""Adjusts points on detected masks based on user input and returns the modified results."""
if self.results[0].masks is not None:
if os.path.isdir(self.source):
raise ValueError(f"'{self.source}' is a directory, not a valid source for this function.")
masks = self._format_results(self.results[0], 0)
target_height, target_width = self.results[0].orig_shape
h = masks[0]["segmentation"].shape[0]
w = masks[0]["segmentation"].shape[1]
if h != target_height or w != target_width:
points = [[int(point[0] * w / target_width), int(point[1] * h / target_height)] for point in points]
onemask = np.zeros((h, w))
for annotation in masks:
mask = annotation["segmentation"] if isinstance(annotation, dict) else annotation
for i, point in enumerate(points):
if mask[point[1], point[0]] == 1 and pointlabel[i] == 1:
onemask += mask
if mask[point[1], point[0]] == 1 and pointlabel[i] == 0:
onemask -= mask
onemask = onemask >= 1
self.results[0].masks.data = torch.tensor(np.array([onemask]))
return self.results
def text_prompt(self, text):
"""Processes a text prompt, applies it to existing results and returns the updated results."""
if self.results[0].masks is not None:
format_results = self._format_results(self.results[0], 0)
cropped_boxes, cropped_images, not_crop, filter_id, annotations = self._crop_image(format_results)
clip_model, preprocess = self.clip.load("ViT-B/32", device=self.device)
scores = self.retrieve(clip_model, preprocess, cropped_boxes, text, device=self.device)
max_idx = scores.argsort()
max_idx = max_idx[-1]
max_idx += sum(np.array(filter_id) <= int(max_idx))
self.results[0].masks.data = torch.tensor(np.array([annotations[max_idx]["segmentation"]]))
return self.results
def everything_prompt(self):
"""Returns the processed results from the previous methods in the class."""
return self.results

67
ultralytics/models/fastsam/utils.py
View File

@ -1,67 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import torch
def adjust_bboxes_to_image_border(boxes, image_shape, threshold=20):
"""
Adjust bounding boxes to stick to image border if they are within a certain threshold.
Args:
boxes (torch.Tensor): (n, 4)
image_shape (tuple): (height, width)
threshold (int): pixel threshold
Returns:
adjusted_boxes (torch.Tensor): adjusted bounding boxes
"""
# Image dimensions
h, w = image_shape
# Adjust boxes
boxes[boxes[:, 0] < threshold, 0] = 0 # x1
boxes[boxes[:, 1] < threshold, 1] = 0 # y1
boxes[boxes[:, 2] > w - threshold, 2] = w # x2
boxes[boxes[:, 3] > h - threshold, 3] = h # y2
return boxes
def bbox_iou(box1, boxes, iou_thres=0.9, image_shape=(640, 640), raw_output=False):
"""
Compute the Intersection-Over-Union of a bounding box with respect to an array of other bounding boxes.
Args:
box1 (torch.Tensor): (4, )
boxes (torch.Tensor): (n, 4)
iou_thres (float): IoU threshold
image_shape (tuple): (height, width)
raw_output (bool): If True, return the raw IoU values instead of the indices
Returns:
high_iou_indices (torch.Tensor): Indices of boxes with IoU > thres
"""
boxes = adjust_bboxes_to_image_border(boxes, image_shape)
# Obtain coordinates for intersections
x1 = torch.max(box1[0], boxes[:, 0])
y1 = torch.max(box1[1], boxes[:, 1])
x2 = torch.min(box1[2], boxes[:, 2])
y2 = torch.min(box1[3], boxes[:, 3])
# Compute the area of intersection
intersection = (x2 - x1).clamp(0) * (y2 - y1).clamp(0)
# Compute the area of both individual boxes
box1_area = (box1[2] - box1[0]) * (box1[3] - box1[1])
box2_area = (boxes[:, 2] - boxes[:, 0]) * (boxes[:, 3] - boxes[:, 1])
# Compute the area of union
union = box1_area + box2_area - intersection
# Compute the IoU
iou = intersection / union # Should be shape (n, )
if raw_output:
return 0 if iou.numel() == 0 else iou
# return indices of boxes with IoU > thres
return torch.nonzero(iou > iou_thres).flatten()

40
ultralytics/models/fastsam/val.py
View File

@ -1,40 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
from ultralytics.models.yolo.segment import SegmentationValidator
from ultralytics.utils.metrics import SegmentMetrics
class FastSAMValidator(SegmentationValidator):
"""
Custom validation class for fast SAM (Segment Anything Model) segmentation in Ultralytics YOLO framework.
Extends the SegmentationValidator class, customizing the validation process specifically for fast SAM. This class
sets the task to 'segment' and uses the SegmentMetrics for evaluation. Additionally, plotting features are disabled
to avoid errors during validation.
Attributes:
dataloader: The data loader object used for validation.
save_dir (str): The directory where validation results will be saved.
pbar: A progress bar object.
args: Additional arguments for customization.
_callbacks: List of callback functions to be invoked during validation.
"""
def __init__(self, dataloader=None, save_dir=None, pbar=None, args=None, _callbacks=None):
"""
Initialize the FastSAMValidator class, setting the task to 'segment' and metrics to SegmentMetrics.
Args:
dataloader (torch.utils.data.DataLoader): Dataloader to be used for validation.
save_dir (Path, optional): Directory to save results.
pbar (tqdm.tqdm): Progress bar for displaying progress.
args (SimpleNamespace): Configuration for the validator.
_callbacks (dict): Dictionary to store various callback functions.
Notes:
Plots for ConfusionMatrix and other related metrics are disabled in this class to avoid errors.
"""
super().__init__(dataloader, save_dir, pbar, args, _callbacks)
self.args.task = "segment"
self.args.plots = False # disable ConfusionMatrix and other plots to avoid errors
self.metrics = SegmentMetrics(save_dir=self.save_dir, on_plot=self.on_plot)

7
ultralytics/models/nas/__init__.py
View File

@ -1,7 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
from .model import NAS
from .predict import NASPredictor
from .val import NASValidator
__all__ = "NASPredictor", "NASValidator", "NAS"

83
ultralytics/models/nas/model.py
View File

@ -1,83 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
"""
YOLO-NAS model interface.
Example:
```python
from ultralytics import NAS
model = NAS('yolo_nas_s')
results = model.predict('ultralytics/assets/bus.jpg')
```
"""
from pathlib import Path
import torch
from ultralytics.engine.model import Model
from ultralytics.utils.torch_utils import model_info, smart_inference_mode
from .predict import NASPredictor
from .val import NASValidator
class NAS(Model):
"""
YOLO NAS model for object detection.
This class provides an interface for the YOLO-NAS models and extends the `Model` class from Ultralytics engine.
It is designed to facilitate the task of object detection using pre-trained or custom-trained YOLO-NAS models.
Example:
```python
from ultralytics import NAS
model = NAS('yolo_nas_s')
results = model.predict('ultralytics/assets/bus.jpg')
```
Attributes:
model (str): Path to the pre-trained model or model name. Defaults to 'yolo_nas_s.pt'.
Note:
YOLO-NAS models only support pre-trained models. Do not provide YAML configuration files.
"""
def __init__(self, model="yolo_nas_s.pt") -> None:
"""Initializes the NAS model with the provided or default 'yolo_nas_s.pt' model."""
assert Path(model).suffix not in (".yaml", ".yml"), "YOLO-NAS models only support pre-trained models."
super().__init__(model, task="detect")
@smart_inference_mode()
def _load(self, weights: str, task: str):
"""Loads an existing NAS model weights or creates a new NAS model with pretrained weights if not provided."""
import super_gradients
suffix = Path(weights).suffix
if suffix == ".pt":
self.model = torch.load(weights)
elif suffix == "":
self.model = super_gradients.training.models.get(weights, pretrained_weights="coco")
# Standardize model
self.model.fuse = lambda verbose=True: self.model
self.model.stride = torch.tensor([32])
self.model.names = dict(enumerate(self.model._class_names))
self.model.is_fused = lambda: False # for info()
self.model.yaml = {} # for info()
self.model.pt_path = weights # for export()
self.model.task = "detect" # for export()
def info(self, detailed=False, verbose=True):
"""
Logs model info.
Args:
detailed (bool): Show detailed information about model.
verbose (bool): Controls verbosity.
"""
return model_info(self.model, detailed=detailed, verbose=verbose, imgsz=640)
@property
def task_map(self):
"""Returns a dictionary mapping tasks to respective predictor and validator classes."""
return {"detect": {"predictor": NASPredictor, "validator": NASValidator}}

60
ultralytics/models/nas/predict.py
View File

@ -1,60 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import torch
from ultralytics.engine.predictor import BasePredictor
from ultralytics.engine.results import Results
from ultralytics.utils import ops
class NASPredictor(BasePredictor):
"""
Ultralytics YOLO NAS Predictor for object detection.
This class extends the `BasePredictor` from Ultralytics engine and is responsible for post-processing the
raw predictions generated by the YOLO NAS models. It applies operations like non-maximum suppression and
scaling the bounding boxes to fit the original image dimensions.
Attributes:
args (Namespace): Namespace containing various configurations for post-processing.
Example:
```python
from ultralytics import NAS
model = NAS('yolo_nas_s')
predictor = model.predictor
# Assumes that raw_preds, img, orig_imgs are available
results = predictor.postprocess(raw_preds, img, orig_imgs)
```
Note:
Typically, this class is not instantiated directly. It is used internally within the `NAS` class.
"""
def postprocess(self, preds_in, img, orig_imgs):
"""Postprocess predictions and returns a list of Results objects."""
# Cat boxes and class scores
boxes = ops.xyxy2xywh(preds_in[0][0])
preds = torch.cat((boxes, preds_in[0][1]), -1).permute(0, 2, 1)
preds = ops.non_max_suppression(
preds,
self.args.conf,
self.args.iou,
agnostic=self.args.agnostic_nms,
max_det=self.args.max_det,
classes=self.args.classes,
)
if not isinstance(orig_imgs, list): # input images are a torch.Tensor, not a list
orig_imgs = ops.convert_torch2numpy_batch(orig_imgs)
results = []
for i, pred in enumerate(preds):
orig_img = orig_imgs[i]
pred[:, :4] = ops.scale_boxes(img.shape[2:], pred[:, :4], orig_img.shape)
img_path = self.batch[0][i]
results.append(Results(orig_img, path=img_path, names=self.model.names, boxes=pred))
return results

50
ultralytics/models/nas/val.py
View File

@ -1,50 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import torch
from ultralytics.models.yolo.detect import DetectionValidator
from ultralytics.utils import ops
__all__ = ["NASValidator"]
class NASValidator(DetectionValidator):
"""
Ultralytics YOLO NAS Validator for object detection.
Extends `DetectionValidator` from the Ultralytics models package and is designed to post-process the raw predictions
generated by YOLO NAS models. It performs non-maximum suppression to remove overlapping and low-confidence boxes,
ultimately producing the final detections.
Attributes:
args (Namespace): Namespace containing various configurations for post-processing, such as confidence and IoU thresholds.
lb (torch.Tensor): Optional tensor for multilabel NMS.
Example:
```python
from ultralytics import NAS
model = NAS('yolo_nas_s')
validator = model.validator
# Assumes that raw_preds are available
final_preds = validator.postprocess(raw_preds)
```
Note:
This class is generally not instantiated directly but is used internally within the `NAS` class.
"""
def postprocess(self, preds_in):
"""Apply Non-maximum suppression to prediction outputs."""
boxes = ops.xyxy2xywh(preds_in[0][0])
preds = torch.cat((boxes, preds_in[0][1]), -1).permute(0, 2, 1)
return ops.non_max_suppression(
preds,
self.args.conf,
self.args.iou,
labels=self.lb,
multi_label=False,
agnostic=self.args.single_cls,
max_det=self.args.max_det,
max_time_img=0.5,
)

7
ultralytics/models/rtdetr/__init__.py
View File

@ -1,7 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
from .model import RTDETR
from .predict import RTDETRPredictor
from .val import RTDETRValidator
__all__ = "RTDETRPredictor", "RTDETRValidator", "RTDETR"

54
ultralytics/models/rtdetr/model.py
View File

@ -1,54 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
"""
Interface for Baidu's RT-DETR, a Vision Transformer-based real-time object detector. RT-DETR offers real-time
performance and high accuracy, excelling in accelerated backends like CUDA with TensorRT. It features an efficient
hybrid encoder and IoU-aware query selection for enhanced detection accuracy.
For more information on RT-DETR, visit: https://arxiv.org/pdf/2304.08069.pdf
"""
from ultralytics.engine.model import Model
from ultralytics.nn.tasks import RTDETRDetectionModel
from .predict import RTDETRPredictor
from .train import RTDETRTrainer
from .val import RTDETRValidator
class RTDETR(Model):
"""
Interface for Baidu's RT-DETR model. This Vision Transformer-based object detector provides real-time performance
with high accuracy. It supports efficient hybrid encoding, IoU-aware query selection, and adaptable inference speed.
Attributes:
model (str): Path to the pre-trained model. Defaults to 'rtdetr-l.pt'.
"""
def __init__(self, model="rtdetr-l.pt") -> None:
"""
Initializes the RT-DETR model with the given pre-trained model file. Supports .pt and .yaml formats.
Args:
model (str): Path to the pre-trained model. Defaults to 'rtdetr-l.pt'.
Raises:
NotImplementedError: If the model file extension is not 'pt', 'yaml', or 'yml'.
"""
super().__init__(model=model, task="detect")
@property
def task_map(self) -> dict:
"""
Returns a task map for RT-DETR, associating tasks with corresponding Ultralytics classes.
Returns:
dict: A dictionary mapping task names to Ultralytics task classes for the RT-DETR model.
"""
return {
"detect": {
"predictor": RTDETRPredictor,
"validator": RTDETRValidator,
"trainer": RTDETRTrainer,
"model": RTDETRDetectionModel,
}
}

86
ultralytics/models/rtdetr/predict.py
View File

@ -1,86 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import torch
from ultralytics.data.augment import LetterBox
from ultralytics.engine.predictor import BasePredictor
from ultralytics.engine.results import Results
from ultralytics.utils import ops
class RTDETRPredictor(BasePredictor):
"""
RT-DETR (Real-Time Detection Transformer) Predictor extending the BasePredictor class for making predictions using
Baidu's RT-DETR model.
This class leverages the power of Vision Transformers to provide real-time object detection while maintaining
high accuracy. It supports key features like efficient hybrid encoding and IoU-aware query selection.
Example:
```python
from ultralytics.utils import ASSETS
from ultralytics.models.rtdetr import RTDETRPredictor
args = dict(model='rtdetr-l.pt', source=ASSETS)
predictor = RTDETRPredictor(overrides=args)
predictor.predict_cli()
```
Attributes:
imgsz (int): Image size for inference (must be square and scale-filled).
args (dict): Argument overrides for the predictor.
"""
def postprocess(self, preds, img, orig_imgs):
"""
Postprocess the raw predictions from the model to generate bounding boxes and confidence scores.
The method filters detections based on confidence and class if specified in `self.args`.
Args:
preds (list): List of [predictions, extra] from the model.
img (torch.Tensor): Processed input images.
orig_imgs (list or torch.Tensor): Original, unprocessed images.
Returns:
(list[Results]): A list of Results objects containing the post-processed bounding boxes, confidence scores,
and class labels.
"""
if not isinstance(preds, (list, tuple)): # list for PyTorch inference but list[0] Tensor for export inference
preds = [preds, None]
nd = preds[0].shape[-1]
bboxes, scores = preds[0].split((4, nd - 4), dim=-1)
if not isinstance(orig_imgs, list): # input images are a torch.Tensor, not a list
orig_imgs = ops.convert_torch2numpy_batch(orig_imgs)
results = []
for i, bbox in enumerate(bboxes): # (300, 4)
bbox = ops.xywh2xyxy(bbox)
score, cls = scores[i].max(-1, keepdim=True) # (300, 1)
idx = score.squeeze(-1) > self.args.conf # (300, )
if self.args.classes is not None:
idx = (cls == torch.tensor(self.args.classes, device=cls.device)).any(1) & idx
pred = torch.cat([bbox, score, cls], dim=-1)[idx] # filter
orig_img = orig_imgs[i]
oh, ow = orig_img.shape[:2]
pred[..., [0, 2]] *= ow
pred[..., [1, 3]] *= oh
img_path = self.batch[0][i]
results.append(Results(orig_img, path=img_path, names=self.model.names, boxes=pred))
return results
def pre_transform(self, im):
"""
Pre-transforms the input images before feeding them into the model for inference. The input images are
letterboxed to ensure a square aspect ratio and scale-filled. The size must be square(640) and scaleFilled.
Args:
im (list[np.ndarray] |torch.Tensor): Input images of shape (N,3,h,w) for tensor, [(h,w,3) x N] for list.
Returns:
(list): List of pre-transformed images ready for model inference.
"""
letterbox = LetterBox(self.imgsz, auto=False, scaleFill=True)
return [letterbox(image=x) for x in im]

101
ultralytics/models/rtdetr/train.py
View File

@ -1,101 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
from copy import copy
import torch
from ultralytics.models.yolo.detect import DetectionTrainer
from ultralytics.nn.tasks import RTDETRDetectionModel
from ultralytics.utils import RANK, colorstr
from .val import RTDETRDataset, RTDETRValidator
class RTDETRTrainer(DetectionTrainer):
"""
Trainer class for the RT-DETR model developed by Baidu for real-time object detection. Extends the DetectionTrainer
class for YOLO to adapt to the specific features and architecture of RT-DETR. This model leverages Vision
Transformers and has capabilities like IoU-aware query selection and adaptable inference speed.
Notes:
- F.grid_sample used in RT-DETR does not support the `deterministic=True` argument.
- AMP training can lead to NaN outputs and may produce errors during bipartite graph matching.
Example:
```python
from ultralytics.models.rtdetr.train import RTDETRTrainer
args = dict(model='rtdetr-l.yaml', data='coco8.yaml', imgsz=640, epochs=3)
trainer = RTDETRTrainer(overrides=args)
trainer.train()
```
"""
def get_model(self, cfg=None, weights=None, verbose=True):
"""
Initialize and return an RT-DETR model for object detection tasks.
Args:
cfg (dict, optional): Model configuration. Defaults to None.
weights (str, optional): Path to pre-trained model weights. Defaults to None.
verbose (bool): Verbose logging if True. Defaults to True.
Returns:
(RTDETRDetectionModel): Initialized model.
"""
model = RTDETRDetectionModel(cfg, nc=self.data["nc"], verbose=verbose and RANK == -1)
if weights:
model.load(weights)
return model
def build_dataset(self, img_path, mode="val", batch=None):
"""
Build and return an RT-DETR dataset for training or validation.
Args:
img_path (str): Path to the folder containing images.
mode (str): Dataset mode, either 'train' or 'val'.
batch (int, optional): Batch size for rectangle training. Defaults to None.
Returns:
(RTDETRDataset): Dataset object for the specific mode.
"""
return RTDETRDataset(
img_path=img_path,
imgsz=self.args.imgsz,
batch_size=batch,
augment=mode == "train",
hyp=self.args,
rect=False,
cache=self.args.cache or None,
prefix=colorstr(f"{mode}: "),
data=self.data,
)
def get_validator(self):
"""
Returns a DetectionValidator suitable for RT-DETR model validation.
Returns:
(RTDETRValidator): Validator object for model validation.
"""
self.loss_names = "giou_loss", "cls_loss", "l1_loss"
return RTDETRValidator(self.test_loader, save_dir=self.save_dir, args=copy(self.args))
def preprocess_batch(self, batch):
"""
Preprocess a batch of images. Scales and converts the images to float format.
Args:
batch (dict): Dictionary containing a batch of images, bboxes, and labels.
Returns:
(dict): Preprocessed batch.
"""
batch = super().preprocess_batch(batch)
bs = len(batch["img"])
batch_idx = batch["batch_idx"]
gt_bbox, gt_class = [], []
for i in range(bs):
gt_bbox.append(batch["bboxes"][batch_idx == i].to(batch_idx.device))
gt_class.append(batch["cls"][batch_idx == i].to(device=batch_idx.device, dtype=torch.long))
return batch

135
ultralytics/models/rtdetr/val.py
View File

@ -1,135 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import torch
from ultralytics.data import YOLODataset
from ultralytics.data.augment import Compose, Format, v8_transforms
from ultralytics.models.yolo.detect import DetectionValidator
from ultralytics.utils import colorstr, ops
__all__ = ("RTDETRValidator",) # tuple or list
class RTDETRDataset(YOLODataset):
"""
Real-Time DEtection and TRacking (RT-DETR) dataset class extending the base YOLODataset class.
This specialized dataset class is designed for use with the RT-DETR object detection model and is optimized for
real-time detection and tracking tasks.
"""
def __init__(self, *args, data=None, **kwargs):
"""Initialize the RTDETRDataset class by inheriting from the YOLODataset class."""
super().__init__(*args, data=data, **kwargs)
# NOTE: add stretch version load_image for RTDETR mosaic
def load_image(self, i, rect_mode=False):
"""Loads 1 image from dataset index 'i', returns (im, resized hw)."""
return super().load_image(i=i, rect_mode=rect_mode)
def build_transforms(self, hyp=None):
"""Temporary, only for evaluation."""
if self.augment:
hyp.mosaic = hyp.mosaic if self.augment and not self.rect else 0.0
hyp.mixup = hyp.mixup if self.augment and not self.rect else 0.0
transforms = v8_transforms(self, self.imgsz, hyp, stretch=True)
else:
# transforms = Compose([LetterBox(new_shape=(self.imgsz, self.imgsz), auto=False, scaleFill=True)])
transforms = Compose([])
transforms.append(
Format(
bbox_format="xywh",
normalize=True,
return_mask=self.use_segments,
return_keypoint=self.use_keypoints,
batch_idx=True,
mask_ratio=hyp.mask_ratio,
mask_overlap=hyp.overlap_mask,
)
)
return transforms
class RTDETRValidator(DetectionValidator):
"""
RTDETRValidator extends the DetectionValidator class to provide validation capabilities specifically tailored for
the RT-DETR (Real-Time DETR) object detection model.
The class allows building of an RTDETR-specific dataset for validation, applies Non-maximum suppression for
post-processing, and updates evaluation metrics accordingly.
Example:
```python
from ultralytics.models.rtdetr import RTDETRValidator
args = dict(model='rtdetr-l.pt', data='coco8.yaml')
validator = RTDETRValidator(args=args)
validator()
```
Note:
For further details on the attributes and methods, refer to the parent DetectionValidator class.
"""
def build_dataset(self, img_path, mode="val", batch=None):
"""
Build an RTDETR Dataset.
Args:
img_path (str): Path to the folder containing images.
mode (str): `train` mode or `val` mode, users are able to customize different augmentations for each mode.
batch (int, optional): Size of batches, this is for `rect`. Defaults to None.
"""
return RTDETRDataset(
img_path=img_path,
imgsz=self.args.imgsz,
batch_size=batch,
augment=False, # no augmentation
hyp=self.args,
rect=False, # no rect
cache=self.args.cache or None,
prefix=colorstr(f"{mode}: "),
data=self.data,
)
def postprocess(self, preds):
"""Apply Non-maximum suppression to prediction outputs."""
if not isinstance(preds, (list, tuple)): # list for PyTorch inference but list[0] Tensor for export inference
preds = [preds, None]
bs, _, nd = preds[0].shape
bboxes, scores = preds[0].split((4, nd - 4), dim=-1)
bboxes *= self.args.imgsz
outputs = [torch.zeros((0, 6), device=bboxes.device)] * bs
for i, bbox in enumerate(bboxes): # (300, 4)
bbox = ops.xywh2xyxy(bbox)
score, cls = scores[i].max(-1) # (300, )
# Do not need threshold for evaluation as only got 300 boxes here
# idx = score > self.args.conf
pred = torch.cat([bbox, score[..., None], cls[..., None]], dim=-1) # filter
# Sort by confidence to correctly get internal metrics
pred = pred[score.argsort(descending=True)]
outputs[i] = pred # [idx]
return outputs
def _prepare_batch(self, si, batch):
"""Prepares a batch for training or inference by applying transformations."""
idx = batch["batch_idx"] == si
cls = batch["cls"][idx].squeeze(-1)
bbox = batch["bboxes"][idx]
ori_shape = batch["ori_shape"][si]
imgsz = batch["img"].shape[2:]
ratio_pad = batch["ratio_pad"][si]
if len(cls):
bbox = ops.xywh2xyxy(bbox) # target boxes
bbox[..., [0, 2]] *= ori_shape[1] # native-space pred
bbox[..., [1, 3]] *= ori_shape[0] # native-space pred
return dict(cls=cls, bbox=bbox, ori_shape=ori_shape, imgsz=imgsz, ratio_pad=ratio_pad)
def _prepare_pred(self, pred, pbatch):
"""Prepares and returns a batch with transformed bounding boxes and class labels."""
predn = pred.clone()
predn[..., [0, 2]] *= pbatch["ori_shape"][1] / self.args.imgsz # native-space pred
predn[..., [1, 3]] *= pbatch["ori_shape"][0] / self.args.imgsz # native-space pred
return predn.float()

6
ultralytics/models/sam/__init__.py
View File

@ -1,6 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
from .model import SAM
from .predict import Predictor
__all__ = "SAM", "Predictor" # tuple or list

187
ultralytics/models/sam/amg.py
View File

@ -1,187 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
import math
from itertools import product
from typing import Any, Generator, List, Tuple
import numpy as np
import torch
def is_box_near_crop_edge(
boxes: torch.Tensor, crop_box: List[int], orig_box: List[int], atol: float = 20.0
) -> torch.Tensor:
"""Return a boolean tensor indicating if boxes are near the crop edge."""
crop_box_torch = torch.as_tensor(crop_box, dtype=torch.float, device=boxes.device)
orig_box_torch = torch.as_tensor(orig_box, dtype=torch.float, device=boxes.device)
boxes = uncrop_boxes_xyxy(boxes, crop_box).float()
near_crop_edge = torch.isclose(boxes, crop_box_torch[None, :], atol=atol, rtol=0)
near_image_edge = torch.isclose(boxes, orig_box_torch[None, :], atol=atol, rtol=0)
near_crop_edge = torch.logical_and(near_crop_edge, ~near_image_edge)
return torch.any(near_crop_edge, dim=1)
def batch_iterator(batch_size: int, *args) -> Generator[List[Any], None, None]:
"""Yield batches of data from the input arguments."""
assert args and all(len(a) == len(args[0]) for a in args), "Batched iteration must have same-size inputs."
n_batches = len(args[0]) // batch_size + int(len(args[0]) % batch_size != 0)
for b in range(n_batches):
yield [arg[b * batch_size : (b + 1) * batch_size] for arg in args]
def calculate_stability_score(masks: torch.Tensor, mask_threshold: float, threshold_offset: float) -> torch.Tensor:
"""
Computes the stability score for a batch of masks.
The stability score is the IoU between the binary masks obtained by thresholding the predicted mask logits at high
and low values.
Notes:
- One mask is always contained inside the other.
- Save memory by preventing unnecessary cast to torch.int64
"""
intersections = (masks > (mask_threshold + threshold_offset)).sum(-1, dtype=torch.int16).sum(-1, dtype=torch.int32)
unions = (masks > (mask_threshold - threshold_offset)).sum(-1, dtype=torch.int16).sum(-1, dtype=torch.int32)
return intersections / unions
def build_point_grid(n_per_side: int) -> np.ndarray:
"""Generate a 2D grid of evenly spaced points in the range [0,1]x[0,1]."""
offset = 1 / (2 * n_per_side)
points_one_side = np.linspace(offset, 1 - offset, n_per_side)
points_x = np.tile(points_one_side[None, :], (n_per_side, 1))
points_y = np.tile(points_one_side[:, None], (1, n_per_side))
return np.stack([points_x, points_y], axis=-1).reshape(-1, 2)
def build_all_layer_point_grids(n_per_side: int, n_layers: int, scale_per_layer: int) -> List[np.ndarray]:
"""Generate point grids for all crop layers."""
return [build_point_grid(int(n_per_side / (scale_per_layer**i))) for i in range(n_layers + 1)]
def generate_crop_boxes(
im_size: Tuple[int, ...], n_layers: int, overlap_ratio: float
) -> Tuple[List[List[int]], List[int]]:
"""
Generates a list of crop boxes of different sizes.
Each layer has (2**i)**2 boxes for the ith layer.
"""
crop_boxes, layer_idxs = [], []
im_h, im_w = im_size
short_side = min(im_h, im_w)
# Original image
crop_boxes.append([0, 0, im_w, im_h])
layer_idxs.append(0)
def crop_len(orig_len, n_crops, overlap):
"""Crops bounding boxes to the size of the input image."""
return int(math.ceil((overlap * (n_crops - 1) + orig_len) / n_crops))
for i_layer in range(n_layers):
n_crops_per_side = 2 ** (i_layer + 1)
overlap = int(overlap_ratio * short_side * (2 / n_crops_per_side))
crop_w = crop_len(im_w, n_crops_per_side, overlap)
crop_h = crop_len(im_h, n_crops_per_side, overlap)
crop_box_x0 = [int((crop_w - overlap) * i) for i in range(n_crops_per_side)]
crop_box_y0 = [int((crop_h - overlap) * i) for i in range(n_crops_per_side)]
# Crops in XYWH format
for x0, y0 in product(crop_box_x0, crop_box_y0):
box = [x0, y0, min(x0 + crop_w, im_w), min(y0 + crop_h, im_h)]
crop_boxes.append(box)
layer_idxs.append(i_layer + 1)
return crop_boxes, layer_idxs
def uncrop_boxes_xyxy(boxes: torch.Tensor, crop_box: List[int]) -> torch.Tensor:
"""Uncrop bounding boxes by adding the crop box offset."""
x0, y0, _, _ = crop_box
offset = torch.tensor([[x0, y0, x0, y0]], device=boxes.device)
# Check if boxes has a channel dimension
if len(boxes.shape) == 3:
offset = offset.unsqueeze(1)
return boxes + offset
def uncrop_points(points: torch.Tensor, crop_box: List[int]) -> torch.Tensor:
"""Uncrop points by adding the crop box offset."""
x0, y0, _, _ = crop_box
offset = torch.tensor([[x0, y0]], device=points.device)
# Check if points has a channel dimension
if len(points.shape) == 3:
offset = offset.unsqueeze(1)
return points + offset
def uncrop_masks(masks: torch.Tensor, crop_box: List[int], orig_h: int, orig_w: int) -> torch.Tensor:
"""Uncrop masks by padding them to the original image size."""
x0, y0, x1, y1 = crop_box
if x0 == 0 and y0 == 0 and x1 == orig_w and y1 == orig_h:
return masks
# Coordinate transform masks
pad_x, pad_y = orig_w - (x1 - x0), orig_h - (y1 - y0)
pad = (x0, pad_x - x0, y0, pad_y - y0)
return torch.nn.functional.pad(masks, pad, value=0)
def remove_small_regions(mask: np.ndarray, area_thresh: float, mode: str) -> Tuple[np.ndarray, bool]:
"""Remove small disconnected regions or holes in a mask, returning the mask and a modification indicator."""
import cv2 # type: ignore
assert mode in {"holes", "islands"}
correct_holes = mode == "holes"
working_mask = (correct_holes ^ mask).astype(np.uint8)
n_labels, regions, stats, _ = cv2.connectedComponentsWithStats(working_mask, 8)
sizes = stats[:, -1][1:] # Row 0 is background label
small_regions = [i + 1 for i, s in enumerate(sizes) if s < area_thresh]
if not small_regions:
return mask, False
fill_labels = [0] + small_regions
if not correct_holes:
# If every region is below threshold, keep largest
fill_labels = [i for i in range(n_labels) if i not in fill_labels] or [int(np.argmax(sizes)) + 1]
mask = np.isin(regions, fill_labels)
return mask, True
def batched_mask_to_box(masks: torch.Tensor) -> torch.Tensor:
"""
Calculates boxes in XYXY format around masks.
Return [0,0,0,0] for an empty mask. For input shape C1xC2x...xHxW, the output shape is C1xC2x...x4.
"""
# torch.max below raises an error on empty inputs, just skip in this case
if torch.numel(masks) == 0:
return torch.zeros(*masks.shape[:-2], 4, device=masks.device)
# Normalize shape to CxHxW
shape = masks.shape
h, w = shape[-2:]
masks = masks.flatten(0, -3) if len(shape) > 2 else masks.unsqueeze(0)
# Get top and bottom edges
in_height, _ = torch.max(masks, dim=-1)
in_height_coords = in_height * torch.arange(h, device=in_height.device)[None, :]
bottom_edges, _ = torch.max(in_height_coords, dim=-1)
in_height_coords = in_height_coords + h * (~in_height)
top_edges, _ = torch.min(in_height_coords, dim=-1)
# Get left and right edges
in_width, _ = torch.max(masks, dim=-2)
in_width_coords = in_width * torch.arange(w, device=in_width.device)[None, :]
right_edges, _ = torch.max(in_width_coords, dim=-1)
in_width_coords = in_width_coords + w * (~in_width)
left_edges, _ = torch.min(in_width_coords, dim=-1)
# If the mask is empty the right edge will be to the left of the left edge.
# Replace these boxes with [0, 0, 0, 0]
empty_filter = (right_edges < left_edges) | (bottom_edges < top_edges)
out = torch.stack([left_edges, top_edges, right_edges, bottom_edges], dim=-1)
out = out * (~empty_filter).unsqueeze(-1)
# Return to original shape
return out.reshape(*shape[:-2], 4) if len(shape) > 2 else out[0]

160
ultralytics/models/sam/build.py
View File

@ -1,160 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
from functools import partial
import torch
from ultralytics.utils.downloads import attempt_download_asset
from .modules.decoders import MaskDecoder
from .modules.encoders import ImageEncoderViT, PromptEncoder
from .modules.sam import Sam
from .modules.tiny_encoder import TinyViT
from .modules.transformer import TwoWayTransformer
def build_sam_vit_h(checkpoint=None):
"""Build and return a Segment Anything Model (SAM) h-size model."""
return _build_sam(
encoder_embed_dim=1280,
encoder_depth=32,
encoder_num_heads=16,
encoder_global_attn_indexes=[7, 15, 23, 31],
checkpoint=checkpoint,
)
def build_sam_vit_l(checkpoint=None):
"""Build and return a Segment Anything Model (SAM) l-size model."""
return _build_sam(
encoder_embed_dim=1024,
encoder_depth=24,
encoder_num_heads=16,
encoder_global_attn_indexes=[5, 11, 17, 23],
checkpoint=checkpoint,
)
def build_sam_vit_b(checkpoint=None):
"""Build and return a Segment Anything Model (SAM) b-size model."""
return _build_sam(
encoder_embed_dim=768,
encoder_depth=12,
encoder_num_heads=12,
encoder_global_attn_indexes=[2, 5, 8, 11],
checkpoint=checkpoint,
)
def build_mobile_sam(checkpoint=None):
"""Build and return Mobile Segment Anything Model (Mobile-SAM)."""
return _build_sam(
encoder_embed_dim=[64, 128, 160, 320],
encoder_depth=[2, 2, 6, 2],
encoder_num_heads=[2, 4, 5, 10],
encoder_global_attn_indexes=None,
mobile_sam=True,
checkpoint=checkpoint,
)
def _build_sam(
encoder_embed_dim, encoder_depth, encoder_num_heads, encoder_global_attn_indexes, checkpoint=None, mobile_sam=False
):
"""Builds the selected SAM model architecture."""
prompt_embed_dim = 256
image_size = 1024
vit_patch_size = 16
image_embedding_size = image_size // vit_patch_size
image_encoder = (
TinyViT(
img_size=1024,
in_chans=3,
num_classes=1000,
embed_dims=encoder_embed_dim,
depths=encoder_depth,
num_heads=encoder_num_heads,
window_sizes=[7, 7, 14, 7],
mlp_ratio=4.0,
drop_rate=0.0,
drop_path_rate=0.0,
use_checkpoint=False,
mbconv_expand_ratio=4.0,
local_conv_size=3,
layer_lr_decay=0.8,
)
if mobile_sam
else ImageEncoderViT(
depth=encoder_depth,
embed_dim=encoder_embed_dim,
img_size=image_size,
mlp_ratio=4,
norm_layer=partial(torch.nn.LayerNorm, eps=1e-6),
num_heads=encoder_num_heads,
patch_size=vit_patch_size,
qkv_bias=True,
use_rel_pos=True,
global_attn_indexes=encoder_global_attn_indexes,
window_size=14,
out_chans=prompt_embed_dim,
)
)
sam = Sam(
image_encoder=image_encoder,
prompt_encoder=PromptEncoder(
embed_dim=prompt_embed_dim,
image_embedding_size=(image_embedding_size, image_embedding_size),
input_image_size=(image_size, image_size),
mask_in_chans=16,
),
mask_decoder=MaskDecoder(
num_multimask_outputs=3,
transformer=TwoWayTransformer(
depth=2,
embedding_dim=prompt_embed_dim,
mlp_dim=2048,
num_heads=8,
),
transformer_dim=prompt_embed_dim,
iou_head_depth=3,
iou_head_hidden_dim=256,
),
pixel_mean=[123.675, 116.28, 103.53],
pixel_std=[58.395, 57.12, 57.375],
)
if checkpoint is not None:
checkpoint = attempt_download_asset(checkpoint)
with open(checkpoint, "rb") as f:
state_dict = torch.load(f)
sam.load_state_dict(state_dict)
sam.eval()
# sam.load_state_dict(torch.load(checkpoint), strict=True)
# sam.eval()
return sam
sam_model_map = {
"sam_h.pt": build_sam_vit_h,
"sam_l.pt": build_sam_vit_l,
"sam_b.pt": build_sam_vit_b,
"mobile_sam.pt": build_mobile_sam,
}
def build_sam(ckpt="sam_b.pt"):
"""Build a SAM model specified by ckpt."""
model_builder = None
ckpt = str(ckpt) # to allow Path ckpt types
for k in sam_model_map.keys():
if ckpt.endswith(k):
model_builder = sam_model_map.get(k)
if not model_builder:
raise FileNotFoundError(f"{ckpt} is not a supported SAM model. Available models are: \n {sam_model_map.keys()}")
return model_builder(ckpt)

114
ultralytics/models/sam/model.py
View File

@ -1,114 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
"""
SAM model interface.
This module provides an interface to the Segment Anything Model (SAM) from Ultralytics, designed for real-time image
segmentation tasks. The SAM model allows for promptable segmentation with unparalleled versatility in image analysis,
and has been trained on the SA-1B dataset. It features zero-shot performance capabilities, enabling it to adapt to new
image distributions and tasks without prior knowledge.
Key Features:
- Promptable segmentation
- Real-time performance
- Zero-shot transfer capabilities
- Trained on SA-1B dataset
"""
from pathlib import Path
from ultralytics.engine.model import Model
from ultralytics.utils.torch_utils import model_info
from .build import build_sam
from .predict import Predictor
class SAM(Model):
"""
SAM (Segment Anything Model) interface class.
SAM is designed for promptable real-time image segmentation. It can be used with a variety of prompts such as
bounding boxes, points, or labels. The model has capabilities for zero-shot performance and is trained on the SA-1B
dataset.
"""
def __init__(self, model="sam_b.pt") -> None:
"""
Initializes the SAM model with a pre-trained model file.
Args:
model (str): Path to the pre-trained SAM model file. File should have a .pt or .pth extension.
Raises:
NotImplementedError: If the model file extension is not .pt or .pth.
"""
if model and Path(model).suffix not in (".pt", ".pth"):
raise NotImplementedError("SAM prediction requires pre-trained *.pt or *.pth model.")
super().__init__(model=model, task="segment")
def _load(self, weights: str, task=None):
"""
Loads the specified weights into the SAM model.
Args:
weights (str): Path to the weights file.
task (str, optional): Task name. Defaults to None.
"""
self.model = build_sam(weights)
def predict(self, source, stream=False, bboxes=None, points=None, labels=None, **kwargs):
"""
Performs segmentation prediction on the given image or video source.
Args:
source (str): Path to the image or video file, or a PIL.Image object, or a numpy.ndarray object.
stream (bool, optional): If True, enables real-time streaming. Defaults to False.
bboxes (list, optional): List of bounding box coordinates for prompted segmentation. Defaults to None.
points (list, optional): List of points for prompted segmentation. Defaults to None.
labels (list, optional): List of labels for prompted segmentation. Defaults to None.
Returns:
(list): The model predictions.
"""
overrides = dict(conf=0.25, task="segment", mode="predict", imgsz=1024)
kwargs.update(overrides)
prompts = dict(bboxes=bboxes, points=points, labels=labels)
return super().predict(source, stream, prompts=prompts, **kwargs)
def __call__(self, source=None, stream=False, bboxes=None, points=None, labels=None, **kwargs):
"""
Alias for the 'predict' method.
Args:
source (str): Path to the image or video file, or a PIL.Image object, or a numpy.ndarray object.
stream (bool, optional): If True, enables real-time streaming. Defaults to False.
bboxes (list, optional): List of bounding box coordinates for prompted segmentation. Defaults to None.
points (list, optional): List of points for prompted segmentation. Defaults to None.
labels (list, optional): List of labels for prompted segmentation. Defaults to None.
Returns:
(list): The model predictions.
"""
return self.predict(source, stream, bboxes, points, labels, **kwargs)
def info(self, detailed=False, verbose=True):
"""
Logs information about the SAM model.
Args:
detailed (bool, optional): If True, displays detailed information about the model. Defaults to False.
verbose (bool, optional): If True, displays information on the console. Defaults to True.
Returns:
(tuple): A tuple containing the model's information.
"""
return model_info(self.model, detailed=detailed, verbose=verbose)
@property
def task_map(self):
"""
Provides a mapping from the 'segment' task to its corresponding 'Predictor'.
Returns:
(dict): A dictionary mapping the 'segment' task to its corresponding 'Predictor'.
"""
return {"segment": {"predictor": Predictor}}

1
ultralytics/models/sam/modules/__init__.py
View File

@ -1 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license

190
ultralytics/models/sam/modules/decoders.py
View File

@ -1,190 +0,0 @@
# Ultralytics YOLO 🚀, AGPL-3.0 license
from typing import List, Tuple, Type
import torch
from torch import nn
from torch.nn import functional as F
from ultralytics.nn.modules import LayerNorm2d
class MaskDecoder(nn.Module):
"""
Decoder module for generating masks and their associated quality scores, using a transformer architecture to predict
masks given image and prompt embeddings.
Attributes:
transformer_dim (int): Channel dimension for the transformer module.
transformer (nn.Module): The transformer module used for mask prediction.
num_multimask_outputs (int): Number of masks to predict for disambiguating masks.
iou_token (nn.Embedding): Embedding for the IoU token.
num_mask_tokens (int): Number of mask tokens.
mask_tokens (nn.Embedding): Embedding for the mask tokens.
output_upscaling (nn.Sequential): Neural network sequence for upscaling the output.
output_hypernetworks_mlps (nn.ModuleList): Hypernetwork MLPs for generating masks.
iou_prediction_head (nn.Module): MLP for predicting mask quality.
"""
def __init__(
self,
*,
transformer_dim: int,
transformer: nn.Module,
num_multimask_outputs: int = 3,
activation: Type[nn.Module] = nn.GELU,
iou_head_depth: int = 3,
iou_head_hidden_dim: int = 256,
) -> None:
"""
Predicts masks given an image and prompt embeddings, using a transformer architecture.
Args:
transformer_dim (int): the channel dimension of the transformer module
transformer (nn.Module): the transformer used to predict masks
num_multimask_outputs (int): the number of masks to predict when disambiguating masks
activation (nn.Module): the type of activation to use when upscaling masks
iou_head_depth (int): the depth of the MLP used to predict mask quality
iou_head_hidden_dim (int): the hidden dimension of the MLP used to predict mask quality
"""
super().__init__()
self.transformer_dim = transformer_dim
self.transformer = transformer
self.num_multimask_outputs = num_multimask_outputs
self.iou_token = nn.Embedding(1, transformer_dim)
self.num_mask_tokens = num_multimask_outputs + 1
self.mask_tokens = nn.Embedding(self.num_mask_tokens, transformer_dim)
self.output_upscaling = nn.Sequential(
nn.ConvTranspose2d(transformer_dim, transformer_dim // 4, kernel_size=2, stride=2),
LayerNorm2d(transformer_dim // 4),
activation(),
nn.ConvTranspose2d(transformer_dim // 4, transformer_dim // 8, kernel_size=2, stride=2),
activation(),
)
self.output_hypernetworks_mlps = nn.ModuleList(
[MLP(transformer_dim, transformer_dim, transformer_dim // 8, 3) for _ in range(self.num_mask_tokens)]
)
self.iou_prediction_head = MLP(transformer_dim, iou_head_hidden_dim, self.num_mask_tokens, iou_head_depth)
def forward(
self,
image_embeddings: torch.Tensor,
image_pe: torch.Tensor,
sparse_prompt_embeddings: torch.Tensor,
dense_prompt_embeddings: torch.Tensor,
multimask_output: bool,
) -> Tuple[torch.Tensor, torch.Tensor]:
"""
Predict masks given image and prompt embeddings.
Args:
image_embeddings (torch.Tensor): the embeddings from the image encoder
image_pe (torch.Tensor): positional encoding with the shape of image_embeddings
sparse_prompt_embeddings (torch.Tensor): the embeddings of the points and boxes
dense_prompt_embeddings (torch.Tensor): the embeddings of the mask inputs
multimask_output (bool): Whether to return multiple masks or a single mask.
Returns:
torch.Tensor: batched predicted masks
torch.Tensor: batched predictions of mask quality
"""
masks, iou_pred = self.predict_masks(
image_embeddings=image_embeddings,
image_pe=image_pe,
sparse_prompt_embeddings=sparse_prompt_embeddings,
dense_prompt_embeddings=dense_prompt_embeddings,
)
# Select the correct mask or masks for output
mask_slice = slice(1, None) if multimask_output else slice(0, 1)
masks = masks[:, mask_slice, :, :]
iou_pred = iou_pred[:, mask_slice]
# Prepare output
return masks, iou_pred
def predict_masks(
self,
image_embeddings: torch.Tensor,
image_pe: torch.Tensor,
sparse_prompt_embeddings: torch.Tensor,
dense_prompt_embeddings: torch.Tensor,
) -> Tuple[torch.Tensor, torch.Tensor]:
"""
Predicts masks.
See 'forward' for more details.
"""
# Concatenate output tokens
output_tokens = torch.cat([self.iou_token.weight, self.mask_tokens.weight], dim=0)
output_tokens = output_tokens.unsqueeze(0).expand(sparse_prompt_embeddings.shape[0], -1, -1)
tokens = torch.cat((output_tokens, sparse_prompt_embeddings), dim=1)
# Expand per-image data in batch direction to be per-mask
src = torch.repeat_interleave(image_embeddings, tokens.shape[0], dim=0)
src = src + dense_prompt_embeddings
pos_src = torch.repeat_interleave(image_pe, tokens.shape[0], dim=0)
b, c, h, w = src.shape
# Run the transformer
hs, src = self.transformer(src, pos_src, tokens)
iou_token_out = hs[:, 0, :]
mask_tokens_out = hs[:, 1 : (1 + self.num_mask_tokens), :]
# Upscale mask embeddings and predict masks using the mask tokens
src = src.transpose(1, 2).view(b, c, h, w)
upscaled_embedding = self.output_upscaling(src)
hyper_in_list: List[torch.Tensor] = [
self.output_hypernetworks_mlps[i](mask_tokens_out[:, i, :]) for i in range(self.num_mask_tokens)
]
hyper_in = torch.stack(hyper_in_list, dim=1)
b, c, h, w = upscaled_embedding.shape
masks = (hyper_in @ upscaled_embedding.view(b, c, h * w)).view(b, -1, h, w)
# Generate mask quality predictions
iou_pred = self.iou_prediction_head(iou_token_out)
return masks, iou_pred
class MLP(nn.Module):
"""
MLP (Multi-Layer Perceptron) model lightly adapted from
https://github.com/facebookresearch/MaskFormer/blob/main/mask_former/modeling/transformer/transformer_predictor.py
"""
def __init__(
self,
input_dim: int,
hidden_dim: int,
output_dim: int,
num_layers: int,
sigmoid_output: bool = False,
) -> None:
"""
Initializes the MLP (Multi-Layer Perceptron) model.
Args:
input_dim (int): The dimensionality of the input features.
hidden_dim (int): The dimensionality of the hidden layers.
output_dim (int): The dimensionality of the output layer.
num_layers (int): The number of hidden layers.
sigmoid_output (bool, optional): Apply a sigmoid activation to the output layer. Defaults to False.
"""
super().__init__()
self.num_layers = num_layers
h = [hidden_dim] * (num_layers - 1)
self.layers = nn.ModuleList(nn.Linear(n, k) for n, k in zip([input_dim] + h, h + [output_dim]))
self.sigmoid_output = sigmoid_output
def forward(self, x):
"""Executes feedforward within the neural network module and applies activation."""
for i, layer in enumerate(self.layers):
x = F.relu(layer(x)) if i < self.num_layers - 1 else layer(x)
if self.sigmoid_output:
x = torch.sigmoid(x)
return x

Some files were not shown because too many files have changed in this diff Show More