ImageProcessor - check if input pixel values between 0-255 (#25688)

* Check if pixel values between 0-255 and add doc clarification

* Add missing docstrings

* _is_scale_image -> is_scaled_image

* Spelling is hard

* Tidy up

src/transformers/image_utils.py

@@ -96,6 +96,17 @@ def is_batched(img):
     return False
 
 
+def is_scaled_image(image: np.ndarray) -> bool:
+    """
+    Checks to see whether the pixel values have already been rescaled to [0, 1].
+    """
+    if image.dtype == np.uint8:
+        return False
+
+    # It's possible the image has pixel values in [0, 255] but is of floating type
+    return np.min(image) >= 0 and np.max(image) <= 1
+
+
 def make_list_of_images(images, expected_ndims: int = 3) -> List[ImageInput]:
     """
     Ensure that the input is a list of images. If the input is a single image, it is converted to a list of length 1.

src/transformers/models/beit/image_processing_beit.py

@@ -28,6 +28,7 @@
     ImageInput,
     PILImageResampling,
     infer_channel_dimension_format,
+    is_scaled_image,
     make_list_of_images,
     to_numpy_array,
     valid_images,
@@ -236,6 +237,11 @@ def _preprocess_image(
         """Preprocesses a single image."""
         # All transformations expect numpy arrays.
         image = to_numpy_array(image)
+        if is_scaled_image(image) and do_rescale:
+            logger.warning_once(
+                "It looks like you are trying to rescale already rescaled images. If the input"
+                " images have pixel values between 0 and 1, set `do_rescale=False` to avoid rescaling them again."
+            )
         if input_data_format is None:
             input_data_format = infer_channel_dimension_format(image)
         image = self._preprocess(
@@ -328,7 +334,8 @@ def preprocess(
 
         Args:
             images (`ImageInput`):
-                Image to preprocess.
+                Image to preprocess. Expects a single or batch of images with pixel values ranging from 0 to 255. If
+                passing in images with pixel values between 0 and 1, set `do_rescale=False`.
             do_resize (`bool`, *optional*, defaults to `self.do_resize`):
                 Whether to resize the image.
             size (`Dict[str, int]`, *optional*, defaults to `self.size`):

src/transformers/models/bit/image_processing_bit.py

@@ -32,6 +32,7 @@
     ImageInput,
     PILImageResampling,
     infer_channel_dimension_format,
+    is_scaled_image,
     make_list_of_images,
     to_numpy_array,
     valid_images,
@@ -184,7 +185,8 @@ def preprocess(
 
         Args:
             images (`ImageInput`):
-                Image to preprocess.
+                Image to preprocess. Expects a single or batch of images with pixel values ranging from 0 to 255. If
+                passing in images with pixel values between 0 and 1, set `do_rescale=False`.
             do_resize (`bool`, *optional*, defaults to `self.do_resize`):
                 Whether to resize the image.
             size (`Dict[str, int]`, *optional*, defaults to `self.size`):
@@ -270,6 +272,12 @@ def preprocess(
         # All transformations expect numpy arrays.
         images = [to_numpy_array(image) for image in images]
 
+        if is_scaled_image(images[0]) and do_rescale:
+            logger.warning_once(
+                "It looks like you are trying to rescale already rescaled images. If the input"
+                " images have pixel values between 0 and 1, set `do_rescale=False` to avoid rescaling them again."
+            )
+
         if input_data_format is None:
             # We assume that all images have the same channel dimension format.
             input_data_format = infer_channel_dimension_format(images[0])

src/transformers/models/blip/image_processing_blip.py

@@ -27,6 +27,7 @@
     ImageInput,
     PILImageResampling,
     infer_channel_dimension_format,
+    is_scaled_image,
     make_list_of_images,
     to_numpy_array,
     valid_images,
@@ -176,7 +177,8 @@ def preprocess(
 
         Args:
             images (`ImageInput`):
-                Image to preprocess.
+                Image to preprocess. Expects a single or batch of images with pixel values ranging from 0 to 255. If
+                passing in images with pixel values between 0 and 1, set `do_rescale=False`.
             do_resize (`bool`, *optional*, defaults to `self.do_resize`):
                 Whether to resize the image.
             size (`Dict[str, int]`, *optional*, defaults to `self.size`):
@@ -253,6 +255,12 @@ def preprocess(
         # All transformations expect numpy arrays.
         images = [to_numpy_array(image) for image in images]
 
+        if is_scaled_image(images[0]) and do_rescale:
+            logger.warning_once(
+                "It looks like you are trying to rescale already rescaled images. If the input"
+                " images have pixel values between 0 and 1, set `do_rescale=False` to avoid rescaling them again."
+            )
+
         if input_data_format is None:
             # We assume that all images have the same channel dimension format.
             input_data_format = infer_channel_dimension_format(images[0])

src/transformers/models/bridgetower/image_processing_bridgetower.py

@@ -29,6 +29,7 @@
     get_image_size,
     infer_channel_dimension_format,
     is_batched,
+    is_scaled_image,
     to_numpy_array,
     valid_images,
 )
@@ -387,7 +388,8 @@ def preprocess(
 
         Args:
             images (`ImageInput`):
-                Image to preprocess.
+                Image to preprocess. Expects a single or batch of images with pixel values ranging from 0 to 255. If
+                passing in images with pixel values between 0 and 1, set `do_rescale=False`.
             do_resize (`bool`, *optional*, defaults to `self.do_resize`):
                 Whether to resize the image.
             size (`Dict[str, int]`, *optional*, defaults to `self.size`):
@@ -469,6 +471,12 @@ def preprocess(
         # All transformations expect numpy arrays.
         images = [to_numpy_array(image) for image in images]
 
+        if is_scaled_image(images[0]) and do_rescale:
+            logger.warning_once(
+                "It looks like you are trying to rescale already rescaled images. If the input"
+                " images have pixel values between 0 and 1, set `do_rescale=False` to avoid rescaling them again."
+            )
+
         if do_resize:
             images = [
                 self.resize(

src/transformers/models/chinese_clip/image_processing_chinese_clip.py

@@ -32,6 +32,7 @@
     ImageInput,
     PILImageResampling,
     infer_channel_dimension_format,
+    is_scaled_image,
     make_list_of_images,
     to_numpy_array,
     valid_images,
@@ -182,7 +183,8 @@ def preprocess(
 
         Args:
             images (`ImageInput`):
-                Image to preprocess.
+                Image to preprocess. Expects a single or batch of images with pixel values ranging from 0 to 255. If
+                passing in images with pixel values between 0 and 1, set `do_rescale=False`.
             do_resize (`bool`, *optional*, defaults to `self.do_resize`):
                 Whether to resize the image.
             size (`Dict[str, int]`, *optional*, defaults to `self.size`):
@@ -268,6 +270,12 @@ def preprocess(
         # All transformations expect numpy arrays.
         images = [to_numpy_array(image) for image in images]
 
+        if is_scaled_image(images[0]) and do_rescale:
+            logger.warning_once(
+                "It looks like you are trying to rescale already rescaled images. If the input"
+                " images have pixel values between 0 and 1, set `do_rescale=False` to avoid rescaling them again."
+            )
+
         if input_data_format is None:
             # We assume that all images have the same channel dimension format.
             input_data_format = infer_channel_dimension_format(images[0])

src/transformers/models/clip/image_processing_clip.py

@@ -32,6 +32,7 @@
     ImageInput,
     PILImageResampling,
     infer_channel_dimension_format,
+    is_scaled_image,
     make_list_of_images,
     to_numpy_array,
     valid_images,
@@ -183,7 +184,8 @@ def preprocess(
 
         Args:
             images (`ImageInput`):
-                Image to preprocess.
+                Image to preprocess. Expects a single or batch of images with pixel values ranging from 0 to 255. If
+                passing in images with pixel values between 0 and 1, set `do_rescale=False`.
             do_resize (`bool`, *optional*, defaults to `self.do_resize`):
                 Whether to resize the image.
             size (`Dict[str, int]`, *optional*, defaults to `self.size`):
@@ -269,6 +271,12 @@ def preprocess(
         # All transformations expect numpy arrays.
         images = [to_numpy_array(image) for image in images]
 
+        if is_scaled_image(images[0]) and do_rescale:
+            logger.warning_once(
+                "It looks like you are trying to rescale already rescaled images. If the input"
+                " images have pixel values between 0 and 1, set `do_rescale=False` to avoid rescaling them again."
+            )
+
         if input_data_format is None:
             # We assume that all images have the same channel dimension format.
             input_data_format = infer_channel_dimension_format(images[0])

src/transformers/models/conditional_detr/image_processing_conditional_detr.py

@@ -42,6 +42,7 @@
     PILImageResampling,
     get_image_size,
     infer_channel_dimension_format,
+    is_scaled_image,
     make_list_of_images,
     to_numpy_array,
     valid_coco_detection_annotations,
@@ -1126,7 +1127,8 @@ def preprocess(
 
         Args:
             images (`ImageInput`):
-                Image or batch of images to preprocess.
+                Image or batch of images to preprocess. Expects a single or batch of images with pixel values ranging
+                from 0 to 255. If passing in images with pixel values between 0 and 1, set `do_rescale=False`.
             annotations (`AnnotationType` or `List[AnnotationType]`, *optional*):
                 List of annotations associated with the image or batch of images. If annotation is for object
                 detection, the annotations should be a dictionary with the following keys:
@@ -1259,6 +1261,12 @@ def preprocess(
         # All transformations expect numpy arrays
         images = [to_numpy_array(image) for image in images]
 
+        if is_scaled_image(images[0]) and do_rescale:
+            logger.warning_once(
+                "It looks like you are trying to rescale already rescaled images. If the input"
+                " images have pixel values between 0 and 1, set `do_rescale=False` to avoid rescaling them again."
+            )
+
         if input_data_format is None:
             # We assume that all images have the same channel dimension format.
             input_data_format = infer_channel_dimension_format(images[0])

src/transformers/models/convnext/image_processing_convnext.py

@@ -32,6 +32,7 @@
     ImageInput,
     PILImageResampling,
     infer_channel_dimension_format,
+    is_scaled_image,
     make_list_of_images,
     to_numpy_array,
     valid_images,
@@ -203,7 +204,8 @@ def preprocess(
 
         Args:
             images (`ImageInput`):
-                Image to preprocess.
+                Image to preprocess. Expects a single or batch of images with pixel values ranging from 0 to 255. If
+                passing in images with pixel values between 0 and 1, set `do_rescale=False`.
             do_resize (`bool`, *optional*, defaults to `self.do_resize`):
                 Whether to resize the image.
             size (`Dict[str, int]`, *optional*, defaults to `self.size`):
@@ -280,6 +282,12 @@ def preprocess(
         # All transformations expect numpy arrays.
         images = [to_numpy_array(image) for image in images]
 
+        if is_scaled_image(images[0]) and do_rescale:
+            logger.warning_once(
+                "It looks like you are trying to rescale already rescaled images. If the input"
+                " images have pixel values between 0 and 1, set `do_rescale=False` to avoid rescaling them again."
+            )
+
         if input_data_format is None:
             # We assume that all images have the same channel dimension format.
             input_data_format = infer_channel_dimension_format(images[0])

src/transformers/models/deformable_detr/image_processing_deformable_detr.py

@@ -42,6 +42,7 @@
     PILImageResampling,
     get_image_size,
     infer_channel_dimension_format,
+    is_scaled_image,
     make_list_of_images,
     to_numpy_array,
     valid_coco_detection_annotations,
@@ -1124,7 +1125,8 @@ def preprocess(
 
         Args:
             images (`ImageInput`):
-                Image or batch of images to preprocess.
+                Image or batch of images to preprocess. Expects a single or batch of images with pixel values ranging
+                from 0 to 255. If passing in images with pixel values between 0 and 1, set `do_rescale=False`.
             annotations (`AnnotationType` or `List[AnnotationType]`, *optional*):
                 List of annotations associated with the image or batch of images. If annotation is for object
                 detection, the annotations should be a dictionary with the following keys:
@@ -1257,6 +1259,12 @@ def preprocess(
         # All transformations expect numpy arrays
         images = [to_numpy_array(image) for image in images]
 
+        if is_scaled_image(images[0]) and do_rescale:
+            logger.warning_once(
+                "It looks like you are trying to rescale already rescaled images. If the input"
+                " images have pixel values between 0 and 1, set `do_rescale=False` to avoid rescaling them again."
+            )
+
         if input_data_format is None:
             # We assume that all images have the same channel dimension format.
             input_data_format = infer_channel_dimension_format(images[0])

src/transformers/models/deit/image_processing_deit.py

@@ -27,6 +27,7 @@
     ImageInput,
     PILImageResampling,
     infer_channel_dimension_format,
+    is_scaled_image,
     make_list_of_images,
     to_numpy_array,
     valid_images,
@@ -180,7 +181,8 @@ def preprocess(
 
         Args:
             images (`ImageInput`):
-                Image to preprocess.
+                Image to preprocess. Expects a single or batch of images with pixel values ranging from 0 to 255. If
+                passing in images with pixel values between 0 and 1, set `do_rescale=False`.
             do_resize (`bool`, *optional*, defaults to `self.do_resize`):
                 Whether to resize the image.
             size (`Dict[str, int]`, *optional*, defaults to `self.size`):
@@ -258,6 +260,12 @@ def preprocess(
         # All transformations expect numpy arrays.
         images = [to_numpy_array(image) for image in images]
 
+        if is_scaled_image(images[0]) and do_rescale:
+            logger.warning_once(
+                "It looks like you are trying to rescale already rescaled images. If the input"
+                " images have pixel values between 0 and 1, set `do_rescale=False` to avoid rescaling them again."
+            )
+
         if input_data_format is None:
             # We assume that all images have the same channel dimension format.
             input_data_format = infer_channel_dimension_format(images[0])

src/transformers/models/deta/image_processing_deta.py

@@ -40,6 +40,7 @@
     get_image_size,
     infer_channel_dimension_format,
     is_batched,
+    is_scaled_image,
     to_numpy_array,
     valid_coco_detection_annotations,
     valid_coco_panoptic_annotations,
@@ -796,7 +797,8 @@ def preprocess(
 
         Args:
             images (`ImageInput`):
-                Image or batch of images to preprocess.
+                Image or batch of images to preprocess. Expects a single or batch of images with pixel values ranging
+                from 0 to 255. If passing in images with pixel values between 0 and 1, set `do_rescale=False`.
             annotations (`List[Dict]` or `List[List[Dict]]`, *optional*):
                 List of annotations associated with the image or batch of images. If annotionation is for object
                 detection, the annotations should be a dictionary with the following keys:
@@ -921,6 +923,12 @@ def preprocess(
         # All transformations expect numpy arrays
         images = [to_numpy_array(image) for image in images]
 
+        if is_scaled_image(images[0]) and do_rescale:
+            logger.warning_once(
+                "It looks like you are trying to rescale already rescaled images. If the input"
+                " images have pixel values between 0 and 1, set `do_rescale=False` to avoid rescaling them again."
+            )
+
         if input_data_format is None:
             # We assume that all images have the same channel dimension format.
             input_data_format = infer_channel_dimension_format(images[0])

src/transformers/models/detr/image_processing_detr.py

@@ -41,6 +41,7 @@
     PILImageResampling,
     get_image_size,
     infer_channel_dimension_format,
+    is_scaled_image,
     make_list_of_images,
     to_numpy_array,
     valid_coco_detection_annotations,
@@ -1096,7 +1097,8 @@ def preprocess(
 
         Args:
             images (`ImageInput`):
-                Image or batch of images to preprocess.
+                Image or batch of images to preprocess. Expects a single or batch of images with pixel values ranging
+                from 0 to 255. If passing in images with pixel values between 0 and 1, set `do_rescale=False`.
             annotations (`AnnotationType` or `List[AnnotationType]`, *optional*):
                 List of annotations associated with the image or batch of images. If annotation is for object
                 detection, the annotations should be a dictionary with the following keys:
@@ -1229,6 +1231,12 @@ def preprocess(
         # All transformations expect numpy arrays
         images = [to_numpy_array(image) for image in images]
 
+        if is_scaled_image(images[0]) and do_rescale:
+            logger.warning_once(
+                "It looks like you are trying to rescale already rescaled images. If the input"
+                " images have pixel values between 0 and 1, set `do_rescale=False` to avoid rescaling them again."
+            )
+
         if input_data_format is None:
             # We assume that all images have the same channel dimension format.
             input_data_format = infer_channel_dimension_format(images[0])