BUGFIX: Dataloading CPU bottleneck (#11195)

Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>
Co-authored-by: UltralyticsAssistant <web@ultralytics.com>

ultralytics/data/build.py

@@ -20,7 +20,7 @@ from ultralytics.data.loaders import (
     autocast_list,
 )
 from ultralytics.data.utils import IMG_FORMATS, VID_FORMATS
-from ultralytics.utils import RANK, colorstr
+from ultralytics.utils import NUM_THREADS, RANK, colorstr
 from ultralytics.utils.checks import check_file
 
 from .dataset import GroundingDataset, YOLODataset, YOLOMultiModalDataset
@@ -81,6 +81,7 @@ def seed_worker(worker_id):  # noqa
     worker_seed = torch.initial_seed() % 2**32
     np.random.seed(worker_seed)
     random.seed(worker_seed)
+    os.sched_setaffinity(0, range(NUM_THREADS))  # fix https://github.com/ultralytics/ultralytics/pull/11195
 
 
 def build_yolo_dataset(cfg, img_path, batch, data, mode="train", rect=False, stride=32, multi_modal=False):

ultralytics/utils/__init__.py

@@ -36,7 +36,7 @@ FILE = Path(__file__).resolve()
 ROOT = FILE.parents[1]  # YOLO
 ASSETS = ROOT / "assets"  # default images
 DEFAULT_CFG_PATH = ROOT / "cfg/default.yaml"
-NUM_THREADS = min(8, max(1, os.cpu_count() - 1))  # number of YOLOv5 multiprocessing threads
+NUM_THREADS = min(8, max(1, os.cpu_count() - 1))  # number of YOLO multiprocessing threads
 AUTOINSTALL = str(os.getenv("YOLO_AUTOINSTALL", True)).lower() == "true"  # global auto-install mode
 VERBOSE = str(os.getenv("YOLO_VERBOSE", True)).lower() == "true"  # global verbose mode
 TQDM_BAR_FORMAT = "{l_bar}{bar:10}{r_bar}" if VERBOSE else None  # tqdm bar format