replace height, width with img_shape

mmcv/transforms/loading.py

@@ -17,10 +17,8 @@ class LoadImageFromFile(BaseTransform):
     Modified Keys:
 
     - img
-    - width
-    - height
-    - ori_width
-    - ori_height
+    - img_shape
+    - ori_shape
 
     Args:
         to_float32 (bool): Whether to convert the loaded image to a float32
@@ -68,11 +66,8 @@ class LoadImageFromFile(BaseTransform):
             img = img.astype(np.float32)
 
         results['img'] = img
-        height, width = img.shape[:2]
-        results['height'] = height
-        results['width'] = width
-        results['ori_height'] = height
-        results['ori_width'] = width
+        results['img_shape'] = img.shape[:2]
+        results['ori_shape'] = img.shape[:2]
         return results
 
     def __repr__(self):
@@ -123,7 +118,7 @@ class LoadAnnotations(BaseTransform):
             # In (x1, y1, x2, y2) order, float type. N is the number of bboxes
             # in np.float32
             'gt_bboxes': np.ndarray(N, 4)
-             # In np.int32 type.
+             # In np.int64 type.
             'gt_bboxes_labels': np.ndarray(N, )
              # In uint8 type.
             'gt_seg_map': np.ndarray (H, W)
@@ -144,7 +139,7 @@ class LoadAnnotations(BaseTransform):
     Added Keys:
 
     - gt_bboxes (np.float32)
-    - gt_bboxes_labels (np.int32)
+    - gt_bboxes_labels (np.int64)
     - gt_seg_map (np.uint8)
     - gt_keypoints (np.float32)
 
@@ -211,7 +206,7 @@ class LoadAnnotations(BaseTransform):
         for instance in results['instances']:
             gt_bboxes_labels.append(instance['bbox_label'])
         results['gt_bboxes_labels'] = np.array(
-            gt_bboxes_labels, dtype=np.int32)
+            gt_bboxes_labels, dtype=np.int64)
 
     def _load_seg_map(self, results: dict) -> None:
         """Private function to load semantic segmentation annotations.

mmcv/transforms/processing.py

@@ -286,8 +286,7 @@ class Pad(BaseTransform):
 
     - img
     - gt_seg_map
-    - height
-    - width
+    - img_shape
 
     Added Keys:
 
@@ -382,8 +381,7 @@ class Pad(BaseTransform):
         results['pad_shape'] = padded_img.shape
         results['pad_fixed_size'] = self.size
         results['pad_size_divisor'] = self.size_divisor
-        results['height'] = padded_img.shape[0]
-        results['width'] = padded_img.shape[1]
+        results['img_shape'] = padded_img.shape[:2]
 
     def _pad_seg(self, results: dict) -> None:
         """Pad semantic segmentation map according to

tests/test_transforms/test_transforms_loading.py

@@ -18,10 +18,8 @@ class TestLoadImageFromFile:
         assert results['img_path'] == osp.join(data_prefix, 'color.jpg')
         assert results['img'].shape == (300, 400, 3)
         assert results['img'].dtype == np.uint8
-        assert results['height'] == 300
-        assert results['width'] == 400
-        assert results['ori_height'] == 300
-        assert results['ori_width'] == 400
+        assert results['img_shape'] == (300, 400)
+        assert results['ori_shape'] == (300, 400)
         assert repr(transform) == transform.__class__.__name__ + \
             "(to_float32=False, color_type='color', " + \
             "imdecode_backend='cv2', file_client_args={'backend': 'disk'})"
@@ -86,7 +84,7 @@ class TestLoadAnnotations:
         results = transform(copy.deepcopy(self.results))
         assert 'gt_bboxes_labels' in results
         assert (results['gt_bboxes_labels'] == np.array([1, 2])).all()
-        assert results['gt_bboxes_labels'].dtype == np.int32
+        assert results['gt_bboxes_labels'].dtype == np.int64
 
     def test_load_kps(self):
         transform = LoadAnnotations(