New dataset fraction=1.0 argument (#2860)

docs/modes/train.md

@@ -82,6 +82,7 @@ task.
 | `close_mosaic`    | `0`      | (int) disable mosaic augmentation for final epochs                          |
 | `resume`          | `False`  | resume training from last checkpoint                                        |
 | `amp`             | `True`   | Automatic Mixed Precision (AMP) training, choices=[True, False]             |
+| `fraction`        | `1.0`    | dataset fraction to train on (default is 1.0, all images in train set)      |
 | `lr0`             | `0.01`   | initial learning rate (i.e. SGD=1E-2, Adam=1E-3)                            |
 | `lrf`             | `0.01`   | final learning rate (lr0 * lrf)                                             |
 | `momentum`        | `0.937`  | SGD momentum/Adam beta1                                                     |

docs/usage/cfg.md

@@ -104,6 +104,7 @@ The training settings for YOLO models encompass various hyperparameters and conf
 | `close_mosaic`    | `0`      | (int) disable mosaic augmentation for final epochs                          |
 | `resume`          | `False`  | resume training from last checkpoint                                        |
 | `amp`             | `True`   | Automatic Mixed Precision (AMP) training, choices=[True, False]             |
+| `fraction`        | `1.0`    | dataset fraction to train on (default is 1.0, all images in train set)      |
 | `lr0`             | `0.01`   | initial learning rate (i.e. SGD=1E-2, Adam=1E-3)                            |
 | `lrf`             | `0.01`   | final learning rate (lr0 * lrf)                                             |
 | `momentum`        | `0.937`  | SGD momentum/Adam beta1                                                     |

ultralytics/yolo/cfg/__init__.py

@@ -66,7 +66,7 @@ CLI_HELP_MSG = \
 CFG_FLOAT_KEYS = 'warmup_epochs', 'box', 'cls', 'dfl', 'degrees', 'shear'
 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', 'mosaic', 'mixup', 'copy_paste', 'conf', 'iou')  # fractional floats limited to 0.0 - 1.0
+                     'fliplr', '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',

ultralytics/yolo/cfg/default.yaml

@@ -30,6 +30,7 @@ cos_lr: False  # use cosine learning rate scheduler
 close_mosaic: 0  # (int) disable mosaic augmentation for final epochs
 resume: False  # resume training from last checkpoint
 amp: True  # Automatic Mixed Precision (AMP) training, choices=[True, False], True runs AMP check
+fraction: 1.0  # dataset fraction to train on (default is 1.0, all images in train set)
 # Segmentation
 overlap_mask: True  # masks should overlap during training (segment train only)
 mask_ratio: 4  # mask downsample ratio (segment train only)

ultralytics/yolo/data/base.py

@@ -36,6 +36,7 @@ class BaseDataset(Dataset):
         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.
@@ -58,13 +59,15 @@ class BaseDataset(Dataset):
                  stride=32,
                  pad=0.5,
                  single_cls=False,
-                 classes=None):
+                 classes=None,
+                 fraction=1.0):
         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
@@ -114,6 +117,8 @@ class BaseDataset(Dataset):
             assert im_files, f'{self.prefix}No images found'
         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)]
         return im_files
 
     def update_labels(self, include_class: Optional[list]):

ultralytics/yolo/data/build.py

@@ -69,7 +69,7 @@ def seed_worker(worker_id):  # noqa
     random.seed(worker_seed)
 
 
-def build_yolo_dataset(cfg, img_path, batch, data_info, mode='train', rect=False, stride=32):
+def build_yolo_dataset(cfg, img_path, batch, data, mode='train', rect=False, stride=32):
     """Build YOLO Dataset"""
     return YOLODataset(
         img_path=img_path,
@@ -86,7 +86,8 @@ def build_yolo_dataset(cfg, img_path, batch, data_info, mode='train', rect=False
         use_segments=cfg.task == 'segment',
         use_keypoints=cfg.task == 'pose',
         classes=cfg.classes,
-        data=data_info)
+        data=data,
+        fraction=cfg.fraction if mode == 'train' else 1.0)
 
 
 def build_dataloader(dataset, batch, workers, shuffle=True, rank=-1):

ultralytics/yolo/data/dataset.py

@@ -226,6 +226,8 @@ class ClassificationDataset(torchvision.datasets.ImageFolder):
             cache (Union[bool, str], optional): Cache setting, can be True, False, 'ram' or 'disk'. Defaults to False.
         """
         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.cache_ram = cache is True or cache == 'ram'
         self.cache_disk = cache == 'disk'
         self.samples = [list(x) + [Path(x[0]).with_suffix('.npy'), None] for x in self.samples]  # file, index, npy, im
@@ -269,4 +271,4 @@ class SemanticDataset(BaseDataset):
 
     def __init__(self):
         """Initialize a SemanticDataset object."""
-        pass
+        super().__init__()