Release 8.0.4 fixes (#256)

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>
Co-authored-by: Laughing <61612323+Laughing-q@users.noreply.github.com>
Co-authored-by: TechieG <35962141+gokulnath30@users.noreply.github.com>
Co-authored-by: Parthiban Marimuthu <66585214+partheee@users.noreply.github.com>

ultralytics/nn/autobackend.py

@@ -56,6 +56,7 @@ class AutoBackend(nn.Module):
         fp16 &= pt or jit or onnx or engine or nn_module  # FP16
         nhwc = coreml or saved_model or pb or tflite or edgetpu  # BHWC formats (vs torch BCWH)
         stride = 32  # default stride
+        model = None  # TODO: resolves ONNX inference, verify effect on other backends
         cuda = torch.cuda.is_available() and device.type != 'cpu'  # use CUDA
         if not (pt or triton or nn_module):
             w = attempt_download(w)  # download if not local

ultralytics/yolo/cli.py

@@ -6,6 +6,7 @@ from pathlib import Path
 import hydra
 
 from ultralytics import hub, yolo
+from ultralytics.yolo.configs import get_config
 from ultralytics.yolo.utils import DEFAULT_CONFIG, LOGGER, colorstr
 
 DIR = Path(__file__).parent
@@ -20,6 +21,9 @@ def cli(cfg):
         cfg (DictConfig): Configuration for the task and mode.
     """
     # LOGGER.info(f"{colorstr(f'Ultralytics YOLO v{ultralytics.__version__}')}")
+    if cfg.cfg:
+        LOGGER.info(f"Overriding default config with {cfg.cfg}")
+        cfg = get_config(cfg.cfg)
     task, mode = cfg.task.lower(), cfg.mode.lower()
 
     # Special case for initializing the configuration
@@ -28,7 +32,7 @@ def cli(cfg):
         LOGGER.info(f"""
         {colorstr("YOLO:")} configuration saved to {Path.cwd() / DEFAULT_CONFIG.name}.
         To run experiments using custom configuration:
-        yolo task='task' mode='mode' --config-name config_file.yaml
+        yolo cfg=config_file.yaml
                     """)
         return
 

ultralytics/yolo/configs/default.yaml

@@ -101,6 +101,7 @@ mixup: 0.0  # image mixup (probability)
 copy_paste: 0.0  # segment copy-paste (probability)
 
 # Hydra configs --------------------------------------------------------------------------------------------------------
+cfg: null # for overriding defaults.yaml
 hydra:
   output_subdir: null  # disable hydra directory creation
   run:

ultralytics/yolo/engine/model.py

@@ -111,7 +111,7 @@ class YOLO:
         self.model.fuse()
 
     @smart_inference_mode()
-    def predict(self, source, **kwargs):
+    def predict(self, source, return_outputs=True, **kwargs):
         """
         Visualize prediction.
 
@@ -127,8 +127,8 @@ class YOLO:
         predictor = self.PredictorClass(overrides=overrides)
 
         predictor.args.imgsz = check_imgsz(predictor.args.imgsz, min_dim=2)  # check image size
-        predictor.setup(model=self.model, source=source)
-        return predictor()
+        predictor.setup(model=self.model, source=source, return_outputs=return_outputs)
+        return predictor() if return_outputs else predictor.predict_cli()
 
     @smart_inference_mode()
     def val(self, data=None, **kwargs):
@@ -212,10 +212,12 @@ class YOLO:
 
     @staticmethod
     def _reset_ckpt_args(args):
-        args.pop("device", None)
         args.pop("project", None)
         args.pop("name", None)
         args.pop("batch", None)
         args.pop("epochs", None)
         args.pop("cache", None)
         args.pop("save_json", None)
+
+        # set device to '' to prevent from auto DDP usage
+        args["device"] = ''

ultralytics/yolo/engine/predictor.py

@@ -89,6 +89,7 @@ class BasePredictor:
         self.vid_path, self.vid_writer = None, None
         self.annotator = None
         self.data_path = None
+        self.output = dict()
         self.callbacks = defaultdict(list, {k: [v] for k, v in callbacks.default_callbacks.items()})  # add callbacks
         callbacks.add_integration_callbacks(self)
 
@@ -104,7 +105,7 @@ class BasePredictor:
     def postprocess(self, preds, img, orig_img):
         return preds
 
-    def setup(self, source=None, model=None):
+    def setup(self, source=None, model=None, return_outputs=True):
         # source
         source = str(source if source is not None else self.args.source)
         is_file = Path(source).suffix[1:] in (IMG_FORMATS + VID_FORMATS)
@@ -155,16 +156,16 @@ class BasePredictor:
         self.imgsz = imgsz
         self.done_setup = True
         self.device = device
+        self.return_outputs = return_outputs
 
         return model
 
     @smart_inference_mode()
-    def __call__(self, source=None, model=None):
+    def __call__(self, source=None, model=None, return_outputs=True):
         self.run_callbacks("on_predict_start")
-        model = self.model if self.done_setup else self.setup(source, model)
+        model = self.model if self.done_setup else self.setup(source, model, return_outputs)
         model.eval()
         self.seen, self.windows, self.dt = 0, [], (ops.Profile(), ops.Profile(), ops.Profile())
-        self.all_outputs = []
         for batch in self.dataset:
             self.run_callbacks("on_predict_batch_start")
             path, im, im0s, vid_cap, s = batch
@@ -194,6 +195,10 @@ class BasePredictor:
                 if self.args.save:
                     self.save_preds(vid_cap, i, str(self.save_dir / p.name))
 
+            if self.return_outputs:
+                yield self.output
+                self.output.clear()
+
             # Print time (inference-only)
             LOGGER.info(f"{s}{'' if len(preds) else '(no detections), '}{self.dt[1].dt * 1E3:.1f}ms")
 
@@ -209,7 +214,11 @@ class BasePredictor:
             LOGGER.info(f"Results saved to {colorstr('bold', self.save_dir)}{s}")
 
         self.run_callbacks("on_predict_end")
-        return self.all_outputs
+
+    def predict_cli(self, source=None, model=None, return_outputs=False):
+        # as __call__ is a genertor now so have to treat it like a genertor
+        for _ in (self.__call__(source, model, return_outputs)):
+            pass
 
     def show(self, p):
         im0 = self.annotator.result()

ultralytics/yolo/utils/torch_utils.py

@@ -70,7 +70,7 @@ def select_device(device='', batch_size=0, newline=False):
     elif device:  # non-cpu device requested
         os.environ['CUDA_VISIBLE_DEVICES'] = device  # set environment variable - must be before assert is_available()
         assert torch.cuda.is_available() and torch.cuda.device_count() >= len(device.replace(',', '')), \
-            f"Invalid CUDA '--device {device}' requested, use '--device cpu' or pass valid CUDA device(s)"
+            f"Invalid CUDA 'device={device}' requested, use 'device=cpu' or pass valid CUDA device(s)"
 
     if not cpu and not mps and torch.cuda.is_available():  # prefer GPU if available
         devices = device.split(',') if device else '0'  # range(torch.cuda.device_count())  # i.e. 0,1,6,7

ultralytics/yolo/v8/classify/predict.py

@@ -39,7 +39,8 @@ class ClassificationPredictor(BasePredictor):
         self.annotator = self.get_annotator(im0)
 
         prob = preds[idx].softmax(0)
-        self.all_outputs.append(prob)
+        if self.return_outputs:
+            self.output["prob"] = prob.cpu().numpy()
         # Print results
         top5i = prob.argsort(0, descending=True)[:5].tolist()  # top 5 indices
         log_string += f"{', '.join(f'{self.model.names[j]} {prob[j]:.2f}' for j in top5i)}, "
@@ -62,7 +63,7 @@ def predict(cfg):
     cfg.source = cfg.source if cfg.source is not None else ROOT / "assets"
 
     predictor = ClassificationPredictor(cfg)
-    predictor()
+    predictor.predict_cli()
 
 
 if __name__ == "__main__":

ultralytics/yolo/v8/classify/train.py

@@ -143,6 +143,7 @@ def train(cfg):
     cfg.weight_decay = 5e-5
     cfg.label_smoothing = 0.1
     cfg.warmup_epochs = 0.0
+    cfg.device = cfg.device if cfg.device is not None else ''
     # trainer = ClassificationTrainer(cfg)
     # trainer.train()
     from ultralytics import YOLO

ultralytics/yolo/v8/detect/predict.py

@@ -53,12 +53,15 @@ class DetectionPredictor(BasePredictor):
         self.annotator = self.get_annotator(im0)
 
         det = preds[idx]
-        self.all_outputs.append(det)
         if len(det) == 0:
             return log_string
         for c in det[:, 5].unique():
             n = (det[:, 5] == c).sum()  # detections per class
             log_string += f"{n} {self.model.names[int(c)]}{'s' * (n > 1)}, "
+
+        if self.return_outputs:
+            self.output["det"] = det.cpu().numpy()
+
         # write
         gn = torch.tensor(im0.shape)[[1, 0, 1, 0]]  # normalization gain whwh
         for *xyxy, conf, cls in reversed(det):
@@ -89,7 +92,7 @@ def predict(cfg):
     cfg.imgsz = check_imgsz(cfg.imgsz, min_dim=2)  # check image size
     cfg.source = cfg.source if cfg.source is not None else ROOT / "assets"
     predictor = DetectionPredictor(cfg)
-    predictor()
+    predictor.predict_cli()
 
 
 if __name__ == "__main__":

ultralytics/yolo/v8/detect/train.py

@@ -199,6 +199,7 @@ class Loss:
 def train(cfg):
     cfg.model = cfg.model or "yolov8n.yaml"
     cfg.data = cfg.data or "coco128.yaml"  # or yolo.ClassificationDataset("mnist")
+    cfg.device = cfg.device if cfg.device is not None else ''
     # trainer = DetectionTrainer(cfg)
     # trainer.train()
     from ultralytics import YOLO

ultralytics/yolo/v8/segment/predict.py

@@ -58,10 +58,10 @@ class SegmentationPredictor(DetectionPredictor):
             return log_string
         # Segments
         mask = masks[idx]
-        if self.args.save_txt:
+        if self.args.save_txt or self.return_outputs:
+            shape = im0.shape if self.args.retina_masks else im.shape[2:]
             segments = [
-                ops.scale_segments(im0.shape if self.args.retina_masks else im.shape[2:], x, im0.shape, normalize=True)
-                for x in reversed(ops.masks2segments(mask))]
+                ops.scale_segments(shape, x, im0.shape, normalize=False) for x in reversed(ops.masks2segments(mask))]
 
         # Print results
         for c in det[:, 5].unique():
@@ -76,12 +76,17 @@ class SegmentationPredictor(DetectionPredictor):
             255 if self.args.retina_masks else im[idx])
 
         det = reversed(det[:, :6])
-        self.all_outputs.append([det, mask])
+        if self.return_outputs:
+            self.output["det"] = det.cpu().numpy()
+            self.output["segment"] = segments
 
         # Write results
-        for j, (*xyxy, conf, cls) in enumerate(reversed(det[:, :6])):
+        for j, (*xyxy, conf, cls) in enumerate(det):
             if self.args.save_txt:  # Write to file
-                seg = segments[j].reshape(-1)  # (n,2) to (n*2)
+                seg = segments[j].copy()
+                seg[:, 0] /= shape[1]  # width
+                seg[:, 1] /= shape[0]  # height
+                seg = seg.reshape(-1)  # (n,2) to (n*2)
                 line = (cls, *seg, conf) if self.args.save_conf else (cls, *seg)  # label format
                 with open(f'{self.txt_path}.txt', 'a') as f:
                     f.write(('%g ' * len(line)).rstrip() % line + '\n')
@@ -106,7 +111,7 @@ def predict(cfg):
     cfg.source = cfg.source if cfg.source is not None else ROOT / "assets"
 
     predictor = SegmentationPredictor(cfg)
-    predictor()
+    predictor.predict_cli()
 
 
 if __name__ == "__main__":

ultralytics/yolo/v8/segment/train.py

@@ -144,6 +144,7 @@ class SegLoss(Loss):
 def train(cfg):
     cfg.model = cfg.model or "yolov8n-seg.yaml"
     cfg.data = cfg.data or "coco128-seg.yaml"  # or yolo.ClassificationDataset("mnist")
+    cfg.device = cfg.device if cfg.device is not None else ''
     # trainer = SegmentationTrainer(cfg)
     # trainer.train()
     from ultralytics import YOLO