ultralytics 8.0.155 allow imgsz and batch resume changes (ultra…

…lytics#4366) Co-authored-by: Mostafa Nemati <[email protected]> Co-authored-by: Eduard Voiculescu <[email protected]>
CV4EcologySchool · Aug 15, 2023 · 9a0555e · 9a0555e
1 parent 60cad0c
commit 9a0555e
Show file tree

Hide file tree

Showing 15 changed files with 84 additions and 105 deletions.
diff --git a/docs/modes/predict.md b/docs/modes/predict.md
@@ -321,7 +321,7 @@ All supported arguments:
 | `augment`      | `bool`         | `False`                | apply image augmentation to prediction sources                                 |
 | `agnostic_nms` | `bool`         | `False`                | class-agnostic NMS                                                             |
 | `retina_masks` | `bool`         | `False`                | use high-resolution segmentation masks                                         |
-| `classes`      | `None or list` | `None`                 | filter results by class, i.e. class=0, or class=[0,2,3]                        |
+| `classes`      | `None or list` | `None`                 | filter results by class, i.e. classes=0, or classes=[0,2,3]                    |
 | `boxes`        | `bool`         | `True`                 | Show boxes in segmentation predictions                                         |
 
 ## Image and Video Formats

diff --git a/docs/reference/trackers/utils/matching.md b/docs/reference/trackers/utils/matching.md
@@ -9,50 +9,18 @@ keywords: Ultralytics, Trackers Utils, Matching, merge_matches, linear_assignmen
 
     Full source code for this file is available at [https://github.com/ultralytics/ultralytics/blob/main/ultralytics/trackers/utils/matching.py](https://github.com/ultralytics/ultralytics/blob/main/ultralytics/trackers/utils/matching.py). Help us fix any issues you see by submitting a [Pull Request](https://docs.ultralytics.com/help/contributing/) 🛠️. Thank you 🙏!
 
----
-## ::: ultralytics.trackers.utils.matching.merge_matches
-<br><br>
-
----
-## ::: ultralytics.trackers.utils.matching._indices_to_matches
-<br><br>
-
 ---
 ## ::: ultralytics.trackers.utils.matching.linear_assignment
 <br><br>
 
----
-## ::: ultralytics.trackers.utils.matching.ious
-<br><br>
-
 ---
 ## ::: ultralytics.trackers.utils.matching.iou_distance
 <br><br>
 
----
-## ::: ultralytics.trackers.utils.matching.v_iou_distance
-<br><br>
-
 ---
 ## ::: ultralytics.trackers.utils.matching.embedding_distance
 <br><br>
 
----
-## ::: ultralytics.trackers.utils.matching.gate_cost_matrix
-<br><br>
-
----
-## ::: ultralytics.trackers.utils.matching.fuse_motion
-<br><br>
-
----
-## ::: ultralytics.trackers.utils.matching.fuse_iou
-<br><br>
-
 ---
 ## ::: ultralytics.trackers.utils.matching.fuse_score
 <br><br>
-
----
-## ::: ultralytics.trackers.utils.matching.bbox_ious
-<br><br>
diff --git a/docs/usage/cfg.md b/docs/usage/cfg.md
@@ -154,7 +154,7 @@ The prediction settings for YOLO models encompass a range of hyperparameters and
 | `augment`      | `False`                | apply image augmentation to prediction sources                                 |
 | `agnostic_nms` | `False`                | class-agnostic NMS                                                             |
 | `retina_masks` | `False`                | use high-resolution segmentation masks                                         |
-| `classes`      | `None`                 | filter results by class, i.e. class=0, or class=[0,2,3]                        |
+| `classes`      | `None`                 | filter results by class, i.e. classes=0, or classes=[0,2,3]                    |
 | `boxes`        | `True`                 | Show boxes in segmentation predictions                                         |
 
 [Predict Guide](../modes/predict.md){ .md-button .md-button--primary}

diff --git a/examples/YOLOv8-CPP-Inference/README.md b/examples/YOLOv8-CPP-Inference/README.md
@@ -8,7 +8,7 @@ This example demonstrates how to perform inference using YOLOv8 and YOLOv5 model
 git clone ultralytics
 cd ultralytics
 pip install .
-cd examples/cpp_
+cd examples/YOLOv8-CPP-Inference
 
 # Add a **yolov8\_.onnx** and/or **yolov5\_.onnx** model(s) to the ultralytics folder.
 # Edit the **main.cpp** to change the **projectBasePath** to match your user.

diff --git a/tests/test_cli.py b/tests/test_cli.py
@@ -55,7 +55,7 @@ def test_predict_online(task, model, data):
     mode = 'track' if task in ('detect', 'segment', 'pose') else 'predict'  # mode for video inference
     model = WEIGHT_DIR / model
     run(f'yolo predict model={model}.pt source=https://ultralytics.com/images/bus.jpg imgsz=32')
-    run(f'yolo {mode} model={model}.pt source=https://ultralytics.com/assets/decelera_landscape_min.mov imgsz=32')
+    run(f'yolo {mode} model={model}.pt source=https://ultralytics.com/assets/decelera_landscape_min.mov imgsz=96')
 
     # Run Python YouTube tracking because CLI is broken. TODO: fix CLI YouTube
     # run(f'yolo {mode} model={model}.pt source=https://youtu.be/G17sBkb38XQ imgsz=32 tracker=bytetrack.yaml')

diff --git a/tests/test_python.py b/tests/test_python.py
@@ -18,6 +18,7 @@
 MODEL = WEIGHTS_DIR / 'path with spaces' / 'yolov8n.pt'  # test spaces in path
 CFG = 'yolov8n.yaml'
 SOURCE = ROOT / 'assets/bus.jpg'
+TMP = (ROOT / '../tests/tmp').resolve()  # temp directory for test files
 SOURCE_GREYSCALE = Path(f'{SOURCE.parent / SOURCE.stem}_greyscale.jpg')
 SOURCE_RGBA = Path(f'{SOURCE.parent / SOURCE.stem}_4ch.png')
 
@@ -92,7 +93,7 @@ def test_predict_grey_and_4ch():
 def test_track_stream():
     # Test YouTube streaming inference (short 10 frame video) with non-default ByteTrack tracker
     model = YOLO(MODEL)
-    model.track('https://youtu.be/G17sBkb38XQ', imgsz=32, tracker='bytetrack.yaml')
+    model.track('https://youtu.be/G17sBkb38XQ', imgsz=96, tracker='bytetrack.yaml')
 
 
 def test_val():
@@ -232,16 +233,15 @@ def test_data_utils():
     # from ultralytics.utils.files import WorkingDirectory
     # with WorkingDirectory(ROOT.parent / 'tests'):
 
-    Path('tests/coco8.zip').unlink(missing_ok=True)
-    Path('coco8.zip').unlink(missing_ok=True)
+    shutil.rmtree(TMP, ignore_errors=True)
+    TMP.mkdir(parents=True)
+
     download('https://github.com/ultralytics/hub/raw/master/example_datasets/coco8.zip', unzip=False)
-    shutil.move('coco8.zip', 'tests')
-    shutil.rmtree('tests/coco8', ignore_errors=True)
-    stats = HUBDatasetStats('tests/coco8.zip', task='detect')
+    shutil.move('coco8.zip', TMP)
+    stats = HUBDatasetStats(TMP / 'coco8.zip', task='detect')
     stats.get_json(save=False)
     stats.process_images()
 
-    autosplit('tests/coco8')
-    zip_directory('tests/coco8/images/val')  # zip
-    shutil.rmtree('tests/coco8', ignore_errors=True)
-    shutil.rmtree('tests/coco8-hub', ignore_errors=True)
+    autosplit(TMP / 'coco8')
+    zip_directory(TMP / 'coco8/images/val')  # zip
+    shutil.rmtree(TMP)
diff --git a/ultralytics/__init__.py b/ultralytics/__init__.py
@@ -1,6 +1,6 @@
 # Ultralytics YOLO 🚀, AGPL-3.0 license
 
-__version__ = '8.0.154'
+__version__ = '8.0.155'
 
 from ultralytics.hub import start
 from ultralytics.models import RTDETR, SAM, YOLO

diff --git a/ultralytics/cfg/default.yaml b/ultralytics/cfg/default.yaml
@@ -64,7 +64,7 @@ line_width:   # (int, optional) line width of the bounding boxes, auto if missin
 visualize: False  # (bool) visualize model features
 augment: False  # (bool) apply image augmentation to prediction sources
 agnostic_nms: False  # (bool) class-agnostic NMS
-classes:  # (int | list[int], optional) filter results by class, i.e. class=0, or class=[0,2,3]
+classes:  # (int | list[int], optional) filter results by class, i.e. classes=0, or classes=[0,2,3]
 retina_masks: False  # (bool) use high-resolution segmentation masks
 boxes: True  # (bool) Show boxes in segmentation predictions
 

diff --git a/ultralytics/data/build.py b/ultralytics/data/build.py
@@ -120,7 +120,7 @@ def check_source(source):
         screenshot = source.lower() == 'screen'
         if is_url and is_file:
             source = check_file(source)  # download
-    elif isinstance(source, tuple(LOADERS)):
+    elif isinstance(source, LOADERS):
         in_memory = True
     elif isinstance(source, (list, tuple)):
         source = autocast_list(source)  # convert all list elements to PIL or np arrays

diff --git a/ultralytics/data/loaders.py b/ultralytics/data/loaders.py
@@ -98,7 +98,7 @@ def update(self, i, cap, stream):
     def close(self):
         """Close stream loader and release resources."""
         self.running = False  # stop flag for Thread
-        for i, thread in enumerate(self.threads):
+        for thread in self.threads:
             if thread.is_alive():
                 thread.join(timeout=5)  # Add timeout
         for cap in self.caps:  # Iterate through the stored VideoCapture objects
@@ -210,7 +210,6 @@ def __init__(self, path, imgsz=640, vid_stride=1):
         self.vid_stride = vid_stride  # video frame-rate stride
         self.bs = 1
         if any(videos):
-            self.orientation = None  # rotation degrees
             self._new_video(videos[0])  # new video
         else:
             self.cap = None
@@ -263,20 +262,6 @@ def _new_video(self, path):
         self.frame = 0
         self.cap = cv2.VideoCapture(path)
         self.frames = int(self.cap.get(cv2.CAP_PROP_FRAME_COUNT) / self.vid_stride)
-        if hasattr(cv2, 'CAP_PROP_ORIENTATION_META'):  # cv2<4.6.0 compatibility
-            self.orientation = int(self.cap.get(cv2.CAP_PROP_ORIENTATION_META))  # rotation degrees
-            # Disable auto-orientation due to known issues in https://github.com/ultralytics/yolov5/issues/8493
-            # self.cap.set(cv2.CAP_PROP_ORIENTATION_AUTO, 0)
-
-    def _cv2_rotate(self, im):
-        """Rotate a cv2 video manually."""
-        if self.orientation == 0:
-            return cv2.rotate(im, cv2.ROTATE_90_CLOCKWISE)
-        elif self.orientation == 180:
-            return cv2.rotate(im, cv2.ROTATE_90_COUNTERCLOCKWISE)
-        elif self.orientation == 90:
-            return cv2.rotate(im, cv2.ROTATE_180)
-        return im
 
     def __len__(self):
         """Returns the number of files in the object."""
@@ -385,10 +370,10 @@ def autocast_list(source):
     return files
 
 
-LOADERS = [LoadStreams, LoadPilAndNumpy, LoadImages, LoadScreenshots]
+LOADERS = LoadStreams, LoadPilAndNumpy, LoadImages, LoadScreenshots  # tuple
 
 
-def get_best_youtube_url(url, use_pafy=True):
+def get_best_youtube_url(url, use_pafy=False):
     """
     Retrieves the URL of the best quality MP4 video stream from a given YouTube video.
 
@@ -411,9 +396,11 @@ def get_best_youtube_url(url, use_pafy=True):
         import yt_dlp
         with yt_dlp.YoutubeDL({'quiet': True}) as ydl:
             info_dict = ydl.extract_info(url, download=False)  # extract info
-        for f in info_dict.get('formats', None):
-            if f['vcodec'] != 'none' and f['acodec'] == 'none' and f['ext'] == 'mp4' and f.get('width') > 1280:
-                return f.get('url', None)
+        for f in reversed(info_dict.get('formats', [])):  # reversed because best is usually last
+            # Find a format with video codec, no audio, *.mp4 extension at least 1920x1080 size
+            good_size = (f.get('width') or 0) >= 1920 or (f.get('height') or 0) >= 1080
+            if good_size and f['vcodec'] != 'none' and f['acodec'] == 'none' and f['ext'] == 'mp4':
+                return f.get('url')
 
 
 if __name__ == '__main__':

diff --git a/ultralytics/data/utils.py b/ultralytics/data/utils.py
@@ -142,16 +142,12 @@ def polygon2mask(imgsz, polygons, color=1, downsample_ratio=1):
         downsample_ratio (int): downsample ratio
     """
     mask = np.zeros(imgsz, dtype=np.uint8)
-    polygons = np.asarray(polygons)
-    polygons = polygons.astype(np.int32)
-    shape = polygons.shape
-    polygons = polygons.reshape(shape[0], -1, 2)
+    polygons = np.asarray(polygons, dtype=np.int32)
+    polygons = polygons.reshape((polygons.shape[0], -1, 2))
     cv2.fillPoly(mask, polygons, color=color)
     nh, nw = (imgsz[0] // downsample_ratio, imgsz[1] // downsample_ratio)
-    # NOTE: fillPoly firstly then resize is trying the keep the same way
-    # of loss calculation when mask-ratio=1.
-    mask = cv2.resize(mask, (nw, nh))
-    return mask
+    # NOTE: fillPoly first then resize is trying to keep the same way of loss calculation when mask-ratio=1.
+    return cv2.resize(mask, (nw, nh))
 
 
 def polygons2masks(imgsz, polygons, color, downsample_ratio=1):
@@ -162,11 +158,7 @@ def polygons2masks(imgsz, polygons, color, downsample_ratio=1):
         color (int): color
         downsample_ratio (int): downsample ratio
     """
-    masks = []
-    for si in range(len(polygons)):
-        mask = polygon2mask(imgsz, [polygons[si].reshape(-1)], color, downsample_ratio)
-        masks.append(mask)
-    return np.array(masks)
+    return np.array([polygon2mask(imgsz, [x.reshape(-1)], color, downsample_ratio) for x in polygons])
 
 
 def polygons2masks_overlap(imgsz, segments, downsample_ratio=1):
@@ -421,7 +413,7 @@ def _round(labels):
             else:
                 raise ValueError('Undefined dataset task.')
             zipped = zip(labels['cls'], coordinates)
-            return [[int(c), *(round(float(x), 4) for x in points)] for c, points in zipped]
+            return [[int(c[0]), *(round(float(x), 4) for x in points)] for c, points in zipped]
 
         for split in 'train', 'val', 'test':
             if self.data.get(split) is None:
@@ -563,7 +555,7 @@ def zip_directory(dir, use_zipfile_library=True):
 
 def autosplit(path=DATASETS_DIR / 'coco8/images', weights=(0.9, 0.1, 0.0), annotated_only=False):
     """
-    Autosplit a dataset into train/val/test splits and save the resulting splits into autosplit_*.txt files.
+    Automatically split a dataset into train/val/test splits and save the resulting splits into autosplit_*.txt files.
 
     Args:
         path (Path, optional): Path to images directory. Defaults to DATASETS_DIR / 'coco8/images'.

diff --git a/ultralytics/engine/exporter.py b/ultralytics/engine/exporter.py
@@ -249,11 +249,11 @@ def __call__(self, model=None):
             f[4], _ = self.export_coreml()
         if any((saved_model, pb, tflite, edgetpu, tfjs)):  # TensorFlow formats
             self.args.int8 |= edgetpu
-            f[5], s_model = self.export_saved_model()
+            f[5], keras_model = self.export_saved_model()
             if pb or tfjs:  # pb prerequisite to tfjs
-                f[6], _ = self.export_pb(s_model)
+                f[6], _ = self.export_pb(keras_model=keras_model)
             if tflite:
-                f[7], _ = self.export_tflite(s_model, nms=False, agnostic_nms=self.args.agnostic_nms)
+                f[7], _ = self.export_tflite(keras_model=keras_model, nms=False, agnostic_nms=self.args.agnostic_nms)
             if edgetpu:
                 f[8], _ = self.export_edgetpu(tflite_model=Path(f[5]) / f'{self.file.stem}_full_integer_quant.tflite')
             if tfjs:
@@ -671,10 +671,7 @@ def export_saved_model(self, prefix=colorstr('TensorFlow SavedModel:')):
         for file in f.rglob('*.tflite'):
             f.unlink() if 'quant_with_int16_act.tflite' in str(f) else self._add_tflite_metadata(file)
 
-        # Load saved_model
-        keras_model = tf.saved_model.load(f, tags=None, options=None)
-
-        return str(f), keras_model
+        return str(f), tf.saved_model.load(f, tags=None, options=None)  # load saved_model as Keras model
 
     @try_export
     def export_pb(self, keras_model, prefix=colorstr('TensorFlow GraphDef:')):

diff --git a/ultralytics/engine/trainer.py b/ultralytics/engine/trainer.py
@@ -81,7 +81,7 @@ def __init__(self, cfg=DEFAULT_CFG, overrides=None, _callbacks=None):
             overrides (dict, optional): Configuration overrides. Defaults to None.
         """
         self.args = get_cfg(cfg, overrides)
-        self.check_resume()
+        self.check_resume(overrides)
         self.device = select_device(self.args.device, self.args.batch)
         self.validator = None
         self.model = None
@@ -576,7 +576,7 @@ def final_eval(self):
                     self.metrics.pop('fitness', None)
                     self.run_callbacks('on_fit_epoch_end')
 
-    def check_resume(self):
+    def check_resume(self, overrides):
         """Check if resume checkpoint exists and update arguments accordingly."""
         resume = self.args.resume
         if resume:
@@ -589,8 +589,13 @@ def check_resume(self):
                 if not Path(ckpt_args['data']).exists():
                     ckpt_args['data'] = self.args.data
 
+                resume = True
                 self.args = get_cfg(ckpt_args)
-                self.args.model, resume = str(last), True  # reinstate
+                self.args.model = str(last)  # reinstate model
+                for k in 'imgsz', 'batch':  # allow arg updates to reduce memory on resume if crashed due to CUDA OOM
+                    if k in overrides:
+                        setattr(self.args, k, overrides[k])
+
             except Exception as e:
                 raise FileNotFoundError('Resume checkpoint not found. Please pass a valid checkpoint to resume from, '
                                         "i.e. 'yolo train resume model=path/to/last.pt'") from e

diff --git a/ultralytics/trackers/utils/__init__.py b/ultralytics/trackers/utils/__init__.py
@@ -0,0 +1 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license