Catch up main -> release/0.12

MANIFEST.in

@@ -0,0 +1,7 @@
+include *.md
+include LICENSE
+include setup.py
+recursive-include yolov5/data *
+recursive-include yolov5/models *
+recursive-include yolov5/models_v5.0 *
+include yolov5/requirements.txt

build_wheel.sh

@@ -0,0 +1,11 @@
+python3 -m pip install --upgrade build
+mkdir yolov5
+cp -r data models models_v5.0 utils .pre-commit-config.yaml $(ls *.py) requirements.txt yolov5/
+grep --include=\*.py -rnl 'yolov5/' -e "from models" | xargs -i@ sed -i 's/from models/from yolov5.models/g' @
+grep --include=\*.py -rnl 'yolov5/' -e "from utils" | xargs -i@ sed -i 's/from utils/from yolov5.utils/g' @
+sed -i '$d' yolov5/requirements.txt
+cat > yolov5/__init__.py << EOF
+from yolov5.utils.general import NM_INTEGRATED
+EOF
+python3 -m build
+rm -r yolov5

export.py

@@ -500,9 +500,9 @@ def load_checkpoint(
     # load sparseml recipe for applying pruning and quantization
     checkpoint_recipe = train_recipe = None
     if resume:
-        train_recipe = ckpt['recipe'] if ('recipe' in ckpt) else None
-    elif ckpt['recipe'] or recipe:
-        train_recipe, checkpoint_recipe = recipe, ckpt['recipe']
+        train_recipe = ckpt.get('recipe')
+    elif recipe or ckpt.get('recipe'):
+        train_recipe, checkpoint_recipe = recipe, ckpt.get('recipe')
 
     sparseml_wrapper = SparseMLWrapper(model.model if val_type else model, checkpoint_recipe, train_recipe)
     exclude_anchors = train_type and (cfg or hyp.get('anchors')) and not resume

pyproject.toml

@@ -0,0 +1,3 @@
+[build-system]
+requires = ["setuptools>=42"]
+build-backend = "setuptools.build_meta"

requirements.txt

@@ -35,4 +35,4 @@ seaborn>=0.11.0
 # pycocotools>=2.0  # COCO mAP
 # roboflow
 thop  # FLOPs computation
-sparseml[torch,torchvision] >= 0.12
+sparseml[torch,torchvision] >= 0.12

setup.py

@@ -0,0 +1,32 @@
+import setuptools
+import os
+
+def read_requirements():
+    build_dir = os.path.abspath(os.path.dirname(__file__))
+    with open(os.path.join(build_dir, "yolov5", "requirements.txt")) as f:
+        return f.read().splitlines()
+
+setuptools.setup(
+    name="yolov5",
+    version='6.1.0',
+    author="",
+    long_description=open("README.md", "r", encoding="utf-8").read(),
+    long_description_content_type="text/markdown",
+    url="https://github.com/ultralytics/yolov5",
+    packages=['yolov5', 'yolov5.models', 'yolov5.utils'],
+    python_requires=">=3.6",
+    install_requires=read_requirements(),
+    include_package_data=True,
+    package_data={'': ['yolov5/models/*.yaml', 'yolov5/data/*']},
+    classifiers=[
+        "Development Status :: 5 - Production/Stable",
+        "License :: OSI Approved :: GNU General Public License (GPL)",
+        "Operating System :: OS Independent",
+        "Intended Audience :: Developers",
+        "Intended Audience :: Science/Research",
+        "Programming Language :: Python :: 3",
+        "Topic :: Software Development :: Libraries :: Python Modules",
+        "Topic :: Education",
+        "Topic :: Scientific/Engineering :: Artificial Intelligence",
+    ]
+)

train.py

@@ -87,9 +87,9 @@ def train(hyp,  # path/to/hyp.yaml or hyp dictionary
     # Save run settings
     if not evolve:
         with open(save_dir / 'hyp.yaml', 'w') as f:
-            yaml.dump(hyp, f, sort_keys=False)
+            yaml.safe_dump(hyp, f, sort_keys=False)
         with open(save_dir / 'opt.yaml', 'w') as f:
-            yaml.dump(vars(opt), f, sort_keys=False)
+            yaml.safe_dump(vars(opt), f, sort_keys=False)
 
     # Loggers
     data_dict = None
@@ -492,7 +492,11 @@ def parse_opt(known=False):
     parser.add_argument('--batch-size', type=int, default=16, help='total batch size for all GPUs, -1 for autobatch')
     parser.add_argument('--imgsz', '--img', '--img-size', type=int, default=640, help='train, val image size (pixels)')
     parser.add_argument('--rect', action='store_true', help='rectangular training')
-    parser.add_argument('--resume', nargs='?', const=True, default=False, help='resume most recent training')
+    parser.add_argument('--resume', 
+                        nargs='?', 
+                        const=True, 
+                        default=False, 
+                        help='resume most recent training. When true, ignores --recipe arg and re-uses saved recipe (if exists)')
     parser.add_argument('--nosave', action='store_true', help='only save final checkpoint')
     parser.add_argument('--noval', action='store_true', help='only validate final epoch')
     parser.add_argument('--noautoanchor', action='store_true', help='disable AutoAnchor')
@@ -542,7 +546,7 @@ def main(opt, callbacks=Callbacks()):
         ckpt = opt.resume if isinstance(opt.resume, str) else get_latest_run()  # specified or most recent path
         assert os.path.isfile(ckpt), 'ERROR: --resume checkpoint does not exist'
         with open(Path(ckpt).parent.parent / 'opt.yaml', errors='ignore') as f:
-            opt = argparse.Namespace(**yaml.load(f, Loader=yaml.SafeLoader))  # replace
+            opt = argparse.Namespace(**yaml.safe_load(f))   # replace
         opt.cfg, opt.weights, opt.resume = '', ckpt, True  # reinstate
         LOGGER.info(f'Resuming training from {ckpt}')
     else:
@@ -553,7 +557,7 @@ def main(opt, callbacks=Callbacks()):
             if opt.project == str(ROOT / 'runs/train'):  # if default project name, rename to runs/evolve
                 opt.project = str(ROOT / 'runs/evolve')
             opt.exist_ok, opt.resume = opt.resume, False  # pass resume to exist_ok and disable resume       
-        opt.save_dir = increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok)  # increment run
+        opt.save_dir = str(increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok)) # increment run
 
     # DDP mode
     device = select_device(opt.device, batch_size=opt.batch_size)

utils/general.py

@@ -40,6 +40,7 @@
 NUM_THREADS = min(8, max(1, os.cpu_count() - 1))  # number of YOLOv5 multiprocessing threads
 VERBOSE = str(os.getenv('YOLOv5_VERBOSE', True)).lower() == 'true'  # global verbose mode
 FONT = 'Arial.ttf'  # https://ultralytics.com/assets/Arial.ttf
+NM_INTEGRATED=True
 
 torch.set_printoptions(linewidth=320, precision=5, profile='long')
 np.set_printoptions(linewidth=320, formatter={'float_kind': '{:11.5g}'.format})  # format short g, %precision=5
@@ -405,7 +406,11 @@ def check_file(file, suffix=''):
         return file
     else:  # search
         files = []
+        if "models_v5.0/" in file:
+            files.extend(glob.glob(str(ROOT / '**' / file), recursive=True))
         for d in 'data', 'models', 'utils':  # search directories
+            if file.startswith(d + os.path.sep):
+                file = file[len(d)+1:]
             files.extend(glob.glob(str(ROOT / d / '**' / file), recursive=True))  # find file
         assert len(files), f'File not found: {file}'  # assert file was found
         assert len(files) == 1, f"Multiple files match '{file}', specify exact path: {files}"  # assert unique
@@ -707,12 +712,13 @@ def clip_coords(boxes, shape):
 
 def non_max_suppression(prediction, conf_thres=0.25, iou_thres=0.45, classes=None, agnostic=False, multi_label=False,
                         labels=(), max_det=300):
-    """Runs Non-Maximum Suppression (NMS) on inference results
+    """Non-Maximum Suppression (NMS) on inference results to reject overlapping bounding boxes
 
     Returns:
          list of detections, on (n,6) tensor per image [xyxy, conf, cls]
     """
 
+    bs = prediction.shape[0]  # batch size
     nc = prediction.shape[2] - 5  # number of classes
     xc = prediction[..., 4] > conf_thres  # candidates
 
@@ -721,18 +727,19 @@ def non_max_suppression(prediction, conf_thres=0.25, iou_thres=0.45, classes=Non
     assert 0 <= iou_thres <= 1, f'Invalid IoU {iou_thres}, valid values are between 0.0 and 1.0'
 
     # Settings
-    min_wh, max_wh = 2, 7680  # (pixels) minimum and maximum box width and height
+    # min_wh = 2  # (pixels) minimum box width and height
+    max_wh = 7680  # (pixels) maximum box width and height
     max_nms = 30000  # maximum number of boxes into torchvision.ops.nms()
-    time_limit = 10.0  # seconds to quit after
+    time_limit = 0.030 * bs  # seconds to quit after
     redundant = True  # require redundant detections
     multi_label &= nc > 1  # multiple labels per box (adds 0.5ms/img)
     merge = False  # use merge-NMS
 
     t = time.time()
-    output = [torch.zeros((0, 6), device=prediction.device)] * prediction.shape[0]
+    output = [torch.zeros((0, 6), device=prediction.device)] * bs
     for xi, x in enumerate(prediction):  # image index, image inference
         # Apply constraints
-        x[((x[..., 2:4] < min_wh) | (x[..., 2:4] > max_wh)).any(1), 4] = 0  # width-height
+        # x[((x[..., 2:4] < min_wh) | (x[..., 2:4] > max_wh)).any(1), 4] = 0  # width-height
         x = x[xc[xi]]  # confidence
 
         # Cat apriori labels if autolabelling
@@ -793,7 +800,7 @@ def non_max_suppression(prediction, conf_thres=0.25, iou_thres=0.45, classes=Non
 
         output[xi] = x[i]
         if (time.time() - t) > time_limit:
-            LOGGER.warning(f'WARNING: NMS time limit {time_limit}s exceeded')
+            LOGGER.warning(f'WARNING: NMS time limit {time_limit:.3f}s exceeded')
             break  # time limit exceeded
 
     return output

utils/sparse.py

@@ -57,9 +57,6 @@ def state_dict(self):
         }
 
     def apply_checkpoint_structure(self):
-        if not self.enabled:
-            return
-
         if self.checkpoint_manager:
             self.checkpoint_manager.apply_structure(self.model, math.inf)