Uses Collection in transformers.image_transforms.normalize (#36301)

* Uses Collection instead of Sequence in transformers.image_transforms.normalize

* Uses collections.abc.Collection in lieu of deprecated typing one

src/transformers/image_transforms.py

@@ -14,8 +14,9 @@
 # limitations under the License.
 
 import warnings
+from collections.abc import Collection
 from math import ceil
-from typing import Dict, Iterable, List, Optional, Sequence, Tuple, Union
+from typing import Dict, Iterable, List, Optional, Tuple, Union
 
 import numpy as np
 
@@ -389,8 +390,8 @@ def resize(
 
 def normalize(
     image: np.ndarray,
-    mean: Union[float, Sequence[float]],
+    mean: Union[float, Collection[float]],
-    std: Union[float, Sequence[float]],
+    std: Union[float, Collection[float]],
     data_format: Optional[ChannelDimension] = None,
     input_data_format: Optional[Union[str, ChannelDimension]] = None,
 ) -> np.ndarray:
@@ -402,9 +403,9 @@ def normalize(
     Args:
         image (`np.ndarray`):
             The image to normalize.
-        mean (`float` or `Sequence[float]`):
+        mean (`float` or `Collection[float]`):
             The mean to use for normalization.
-        std (`float` or `Sequence[float]`):
+        std (`float` or `Collection[float]`):
             The standard deviation to use for normalization.
         data_format (`ChannelDimension`, *optional*):
             The channel dimension format of the output image. If unset, will use the inferred format from the input.
@@ -425,14 +426,14 @@ def normalize(
     if not np.issubdtype(image.dtype, np.floating):
         image = image.astype(np.float32)
 
-    if isinstance(mean, Sequence):
+    if isinstance(mean, Collection):
         if len(mean) != num_channels:
             raise ValueError(f"mean must have {num_channels} elements if it is an iterable, got {len(mean)}")
     else:
         mean = [mean] * num_channels
     mean = np.array(mean, dtype=image.dtype)
 
-    if isinstance(std, Sequence):
+    if isinstance(std, Collection):
         if len(std) != num_channels:
             raise ValueError(f"std must have {num_channels} elements if it is an iterable, got {len(std)}")
     else:

tests/test_image_transforms.py

@@ -326,8 +326,8 @@ class ImageTransformsTester(unittest.TestCase):
 
         # Test float16 image input keeps float16 dtype
         image = np.random.randint(0, 256, (224, 224, 3)).astype(np.float16) / 255
-        mean = (0.5, 0.6, 0.7)
+        mean = np.array((0.5, 0.6, 0.7))
-        std = (0.1, 0.2, 0.3)
+        std = np.array((0.1, 0.2, 0.3))
 
         # The mean and std are cast to match the dtype of the input image
         cast_mean = np.array(mean, dtype=np.float16)