ultralytics 8.0.206 engine Trainer updates (#6111)

Signed-off-by: Glenn Jocher <glenn.jocher@ultralytics.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: jamjamjon <51357717+jamjamjon@users.noreply.github.com>

.github/workflows/links.yml

@@ -1,6 +1,11 @@
 # Ultralytics YOLO 🚀, AGPL-3.0 license
-# YOLO Continuous Integration (CI) GitHub Actions tests broken link checker
+# Continuous Integration (CI) GitHub Actions tests broken link checker using https://github.com/lycheeverse/lychee
-# Accept 429(Instagram, 'too many requests'), 999(LinkedIn, 'unknown status code'), Timeout(Twitter)
+# Ignores the following status codes to reduce false positives:
+#   - 403(OpenVINO, 'forbidden')
+#   - 429(Instagram, 'too many requests')
+#   - 500(Zenodo, 'cached')
+#   - 502(Zenodo, 'bad gateway')
+#   - 999(LinkedIn, 'unknown status code')
 
 name: Check Broken links
 
@@ -28,7 +33,7 @@ jobs:
           timeout_minutes: 5
           retry_wait_seconds: 60
           max_attempts: 3
-          command: lychee --accept 429,999 --exclude-loopback --exclude 'https?://(www\.)?(linkedin\.com|twitter\.com|instagram\.com|kaggle\.com|fonts\.gstatic\.com|url\.com)' --exclude-path '**/ci.yaml' --exclude-mail --github-token ${{ secrets.GITHUB_TOKEN }} './**/*.md' './**/*.html'
+          command: lychee --accept 403,429,500,502,999 --exclude-loopback --exclude 'https?://(www\.)?(linkedin\.com|twitter\.com|instagram\.com|kaggle\.com|fonts\.gstatic\.com|url\.com)' --exclude-path '**/ci.yaml' --exclude-mail --github-token ${{ secrets.GITHUB_TOKEN }} './**/*.md' './**/*.html'
 
       - name: Test Markdown, HTML, YAML, Python and Notebook links with retry
         if: github.event_name == 'workflow_dispatch'
@@ -37,4 +42,4 @@ jobs:
           timeout_minutes: 5
           retry_wait_seconds: 60
           max_attempts: 3
-          command: lychee --accept 429,999 --exclude-loopback --exclude 'https?://(www\.)?(linkedin\.com|twitter\.com|instagram\.com|kaggle\.com|url\.com|fonts\.gstatic\.com|url\.com)' --exclude-path '**/ci.yaml' --exclude-mail --github-token ${{ secrets.GITHUB_TOKEN }} './**/*.md' './**/*.html' './**/*.yml' './**/*.yaml' './**/*.py' './**/*.ipynb'
+          command: lychee --accept 429,999 --exclude-loopback --exclude 'https?://(www\.)?(linkedin\.com|twitter\.com|instagram\.com|kaggle\.com|fonts\.gstatic\.com|url\.com)' --exclude-path '**/ci.yaml' --exclude-mail --github-token ${{ secrets.GITHUB_TOKEN }} './**/*.md' './**/*.html' './**/*.yml' './**/*.yaml' './**/*.py' './**/*.ipynb'

docs/help/privacy.md

@@ -7,7 +7,7 @@ keywords: Ultralytics, Data Collection, User Privacy, Google Analytics, Sentry,
 
 ## Overview
 
-Ultralytics is dedicated to the continuous enhancement of the user experience and the capabilities of our Python package, including the advanced YOLO models we develop. Our approach involves the gathering of anonymized usage statistics and crash reports, helping us identify opportunities for improvement and ensuring the reliability of our software. This transparency document outlines what data we collect, its purpose, and the choice you have regarding this data collection.
+[Ultralytics](https://ultralytics.com) is dedicated to the continuous enhancement of the user experience and the capabilities of our Python package, including the advanced YOLO models we develop. Our approach involves the gathering of anonymized usage statistics and crash reports, helping us identify opportunities for improvement and ensuring the reliability of our software. This transparency document outlines what data we collect, its purpose, and the choice you have regarding this data collection.
 
 ## Anonymized Google Analytics
 
@@ -50,7 +50,7 @@ If the `sentry-sdk` Python package is pre-installed on your system a crash event
 - **Crash Logs**: Detailed reports on the application's condition at the time of a crash, which are vital for our debugging efforts.
 - **Error Messages**: We record error messages generated during the operation of our package to understand and resolve potential issues quickly.
 
-To learn more about how Sentry handles data, please visit [Sentry Privacy Policy](https://sentry.io/privacy/).
+To learn more about how Sentry handles data, please visit [Sentry's Privacy Policy](https://sentry.io/privacy/).
 
 ### How We Use This Data
 

ultralytics/__init__.py

@@ -1,6 +1,6 @@
 # Ultralytics YOLO 🚀, AGPL-3.0 license
 
-__version__ = '8.0.205'
+__version__ = '8.0.206'
 
 from ultralytics.models import RTDETR, SAM, YOLO
 from ultralytics.models.fastsam import FastSAM

ultralytics/engine/model.py

@@ -9,7 +9,6 @@ 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, callbacks, checks, emojis, yaml_load
-from ultralytics.utils.downloads import GITHUB_ASSETS_STEMS
 
 
 class Model(nn.Module):
@@ -88,10 +87,8 @@ class Model(nn.Module):
             return
 
         # Load or create new YOLO model
-        suffix = Path(model).suffix
+        model = checks.check_model_file_from_stem(model)  # add suffix, i.e. yolov8n -> yolov8n.pt
-        if not suffix and Path(model).stem in GITHUB_ASSETS_STEMS:
+        if Path(model).suffix in ('.yaml', '.yml'):
-            model, suffix = Path(model).with_suffix('.pt'), '.pt'  # add suffix, i.e. yolov8n -> yolov8n.pt
-        if suffix in ('.yaml', '.yml'):
             self._new(model, task)
         else:
             self._load(model, task)

ultralytics/engine/trainer.py

@@ -19,8 +19,6 @@ import numpy as np
 import torch
 from torch import distributed as dist
 from torch import nn, optim
-from torch.cuda import amp
-from torch.nn.parallel import DistributedDataParallel as DDP
 
 from ultralytics.cfg import get_cfg, get_save_dir
 from ultralytics.data.utils import check_cls_dataset, check_det_dataset
@@ -28,7 +26,7 @@ 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, print_args
+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,
@@ -43,7 +41,6 @@ class BaseTrainer:
 
     Attributes:
         args (SimpleNamespace): Configuration for the trainer.
-        check_resume (method): Method to check if training should be resumed from a saved checkpoint.
         validator (BaseValidator): Validator instance.
         model (nn.Module): Model instance.
         callbacks (defaultdict): Dictionary of callbacks.
@@ -62,6 +59,7 @@ class BaseTrainer:
         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.
@@ -84,7 +82,6 @@ class BaseTrainer:
         self.check_resume(overrides)
         self.device = select_device(self.args.device, self.args.batch)
         self.validator = None
-        self.model = None
         self.metrics = None
         self.plots = {}
         init_seeds(self.args.seed + 1 + RANK, deterministic=self.args.deterministic)
@@ -111,7 +108,7 @@ class BaseTrainer:
             self.args.workers = 0  # faster CPU training as time dominated by inference, not dataloading
 
         # Model and Dataset
-        self.model = self.args.model
+        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)
@@ -124,6 +121,7 @@ class BaseTrainer:
 
         self.trainset, self.testset = self.get_dataset(self.data)
         self.ema = None
+        self.resume = False
 
         # Optimization utils init
         self.lf = None
@@ -236,9 +234,9 @@ class BaseTrainer:
         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 = amp.GradScaler(enabled=self.amp)
+        self.scaler = torch.cuda.amp.GradScaler(enabled=self.amp)
         if world_size > 1:
-            self.model = DDP(self.model, device_ids=[RANK])
+            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)
@@ -311,11 +309,7 @@ class BaseTrainer:
             pbar = enumerate(self.train_loader)
             # Update dataloader attributes (optional)
             if epoch == (self.epochs - self.args.close_mosaic):
-                LOGGER.info('Closing dataloader mosaic')
+                self._close_dataloader_mosaic()
-                if hasattr(self.train_loader.dataset, 'mosaic'):
-                    self.train_loader.dataset.mosaic = False
-                if hasattr(self.train_loader.dataset, 'close_mosaic'):
-                    self.train_loader.dataset.close_mosaic(hyp=self.args)
                 self.train_loader.reset()
 
             if RANK in (-1, 0):
@@ -395,7 +389,7 @@ class BaseTrainer:
             self.epoch_time = tnow - self.epoch_time_start
             self.epoch_time_start = tnow
             self.run_callbacks('on_fit_epoch_end')
-            torch.cuda.empty_cache()  # clears GPU vRAM at end of epoch, can help with out of memory errors
+            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
@@ -613,11 +607,15 @@ class BaseTrainer:
         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')
-            if hasattr(self.train_loader.dataset, 'mosaic'):
+            self.train_loader.dataset.close_mosaic(hyp=self.args)
-                self.train_loader.dataset.mosaic = False
-            if hasattr(self.train_loader.dataset, 'close_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):
         """

ultralytics/utils/checks.py

@@ -426,6 +426,14 @@ def check_yolov5u_filename(file: str, verbose: bool = True):
     return file
 
 
+def check_model_file_from_stem(model='yolov8n'):
+    """Return a model filename from a valid model stem."""
+    if model and not Path(model).suffix and Path(model).stem in downloads.GITHUB_ASSETS_STEMS:
+        return Path(model).with_suffix('.pt')  # add suffix, i.e. yolov8n -> yolov8n.pt
+    else:
+        return model
+
+
 def check_file(file, suffix='', download=True, hard=True):
     """Search/download file (if necessary) and return path."""
     check_suffix(file, suffix)  # optional

ultralytics/utils/ops.py

@@ -324,8 +324,8 @@ def scale_image(masks, im0_shape, ratio_pad=None):
     else:
         gain = ratio_pad[0][0]
         pad = ratio_pad[1]
-    top, left = int(pad[1]), int(pad[0])  # y, x
+    top, left = (int(round(pad[1] - 0.1)), int(round(pad[0] - 0.1)))  # y, x
-    bottom, right = int(im1_shape[0] - pad[1]), int(im1_shape[1] - pad[0])
+    bottom, right = (int(round(im1_shape[0] - pad[1] + 0.1)), int(round(im1_shape[1] - pad[0] + 0.1)))
 
     if len(masks.shape) < 2:
         raise ValueError(f'"len of masks shape" should be 2 or 3, but got {len(masks.shape)}')
@@ -704,8 +704,8 @@ def scale_masks(masks, shape, padding=True):
     if padding:
         pad[0] /= 2
         pad[1] /= 2
-    top, left = (int(pad[1]), int(pad[0])) if padding else (0, 0)  # y, x
+    top, left = (int(round(pad[1] - 0.1)), int(round(pad[0] - 0.1))) if padding else (0, 0)  # y, x
-    bottom, right = (int(mh - pad[1]), int(mw - pad[0]))
+    bottom, right = (int(round(mh - pad[1] + 0.1)), int(round(mw - pad[0] + 0.1)))
     masks = masks[..., top:bottom, left:right]
 
     masks = F.interpolate(masks, shape, mode='bilinear', align_corners=False)  # NCHW