Add ViTImageProcessorFast to tests (#31424)

* Add ViTImageProcessor to tests

* Correct data format

* Review comments

tests/models/bridgetower/test_image_processing_bridgetower.py

@@ -17,6 +17,8 @@
 import unittest
 from typing import Dict, List, Optional, Union
 
+import numpy as np
+
 from transformers.testing_utils import require_torch, require_vision
 from transformers.utils import is_vision_available
 
@@ -84,6 +86,8 @@ class BridgeTowerImageProcessingTester(unittest.TestCase):
             image = image_inputs[0]
             if isinstance(image, Image.Image):
                 w, h = image.size
+            elif isinstance(image, np.ndarray):
+                h, w = image.shape[0], image.shape[1]
             else:
                 h, w = image.shape[1], image.shape[2]
             scale = size / min(w, h)

tests/models/conditional_detr/test_image_processing_conditional_detr.py

@@ -18,6 +18,8 @@ import json
 import pathlib
 import unittest
 
+import numpy as np
+
 from transformers.testing_utils import require_torch, require_vision, slow
 from transformers.utils import is_torch_available, is_vision_available
 
@@ -87,6 +89,8 @@ class ConditionalDetrImageProcessingTester(unittest.TestCase):
             image = image_inputs[0]
             if isinstance(image, Image.Image):
                 w, h = image.size
+            elif isinstance(image, np.ndarray):
+                h, w = image.shape[0], image.shape[1]
             else:
                 h, w = image.shape[1], image.shape[2]
             if w < h:

tests/models/deformable_detr/test_image_processing_deformable_detr.py

@@ -18,6 +18,8 @@ import json
 import pathlib
 import unittest
 
+import numpy as np
+
 from transformers.testing_utils import require_torch, require_vision, slow
 from transformers.utils import is_torch_available, is_vision_available
 
@@ -87,6 +89,8 @@ class DeformableDetrImageProcessingTester(unittest.TestCase):
             image = image_inputs[0]
             if isinstance(image, Image.Image):
                 w, h = image.size
+            elif isinstance(image, np.ndarray):
+                h, w = image.shape[0], image.shape[1]
             else:
                 h, w = image.shape[1], image.shape[2]
             if w < h:

tests/models/detr/test_image_processing_detr.py

@@ -17,6 +17,8 @@ import json
 import pathlib
 import unittest
 
+import numpy as np
+
 from transformers.testing_utils import require_torch, require_vision, slow
 from transformers.utils import is_torch_available, is_vision_available
 
@@ -86,6 +88,8 @@ class DetrImageProcessingTester(unittest.TestCase):
             image = image_inputs[0]
             if isinstance(image, Image.Image):
                 w, h = image.size
+            elif isinstance(image, np.ndarray):
+                h, w = image.shape[0], image.shape[1]
             else:
                 h, w = image.shape[1], image.shape[2]
             if w < h:

tests/models/glpn/test_image_processing_glpn.py

@@ -66,6 +66,8 @@ class GLPNImageProcessingTester(unittest.TestCase):
     def expected_output_image_shape(self, images):
         if isinstance(images[0], Image.Image):
             width, height = images[0].size
+        elif isinstance(images[0], np.ndarray):
+            height, width = images[0].shape[0], images[0].shape[1]
         else:
             height, width = images[0].shape[1], images[0].shape[2]
 

tests/models/grounding_dino/test_image_processing_grounding_dino.py

@@ -18,6 +18,8 @@ import json
 import pathlib
 import unittest
 
+import numpy as np
+
 from transformers.testing_utils import require_torch, require_vision, slow
 from transformers.utils import is_torch_available, is_vision_available
 
@@ -93,6 +95,8 @@ class GroundingDinoImageProcessingTester(unittest.TestCase):
             image = image_inputs[0]
             if isinstance(image, Image.Image):
                 w, h = image.size
+            elif isinstance(image, np.ndarray):
+                h, w = image.shape[0], image.shape[1]
             else:
                 h, w = image.shape[1], image.shape[2]
             if w < h:

tests/models/idefics/test_image_processing_idefics.py

@@ -16,6 +16,8 @@
 
 import unittest
 
+import numpy as np
+
 from transformers.testing_utils import require_torch, require_torchvision, require_vision
 from transformers.utils import is_torch_available, is_torchvision_available, is_vision_available
 
@@ -75,6 +77,8 @@ class IdeficsImageProcessingTester(unittest.TestCase):
             image = image_inputs[0]
             if isinstance(image, Image.Image):
                 w, h = image.size
+            elif isinstance(image, np.ndarray):
+                h, w = image.shape[0], image.shape[1]
             else:
                 h, w = image.shape[1], image.shape[2]
             scale = size / min(w, h)

tests/models/idefics2/test_image_processing_idefics2.py

@@ -99,6 +99,8 @@ class Idefics2ImageProcessingTester(unittest.TestCase):
             image = image_inputs[0]
             if isinstance(image, Image.Image):
                 w, h = image.size
+            elif isinstance(image, np.ndarray):
+                h, w = image.shape[0], image.shape[1]
             else:
                 h, w = image.shape[1], image.shape[2]
 
@@ -176,6 +178,10 @@ class Idefics2ImageProcessingTester(unittest.TestCase):
         if torchify:
             images_list = [[torch.from_numpy(image) for image in images] for images in images_list]
 
+        if numpify:
+            # Numpy images are typically in channels last format
+            images_list = [[image.transpose(1, 2, 0) for image in images] for images in images_list]
+
         return images_list
 
 
@@ -206,66 +212,100 @@ class Idefics2ImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
         self.assertTrue(hasattr(image_processing, "do_image_splitting"))
 
     def test_call_numpy(self):
-        # Initialize image_processing
-        image_processing = self.image_processing_class(**self.image_processor_dict)
-        # create random numpy tensors
-        image_inputs = self.image_processor_tester.prepare_image_inputs(equal_resolution=False, numpify=True)
-        for sample_images in image_inputs:
-            for image in sample_images:
-                self.assertIsInstance(image, np.ndarray)
+        for image_processing_class in self.image_processor_list:
+            # Initialize image_processing
+            image_processing = self.image_processing_class(**self.image_processor_dict)
+            # create random numpy tensors
+            image_inputs = self.image_processor_tester.prepare_image_inputs(equal_resolution=False, numpify=True)
+            for sample_images in image_inputs:
+                for image in sample_images:
+                    self.assertIsInstance(image, np.ndarray)
 
-        # Test not batched input
-        encoded_images = image_processing(image_inputs[0], return_tensors="pt").pixel_values
-        expected_output_image_shape = self.image_processor_tester.expected_output_image_shape([image_inputs[0]])
-        self.assertEqual(tuple(encoded_images.shape), (1, *expected_output_image_shape))
+            # Test not batched input
+            encoded_images = image_processing(image_inputs[0], return_tensors="pt").pixel_values
+            expected_output_image_shape = self.image_processor_tester.expected_output_image_shape([image_inputs[0]])
+            self.assertEqual(tuple(encoded_images.shape), (1, *expected_output_image_shape))
 
-        # Test batched
-        encoded_images = image_processing(image_inputs, return_tensors="pt").pixel_values
-        expected_output_image_shape = self.image_processor_tester.expected_output_image_shape(image_inputs)
-        self.assertEqual(
-            tuple(encoded_images.shape), (self.image_processor_tester.batch_size, *expected_output_image_shape)
-        )
+            # Test batched
+            encoded_images = image_processing(image_inputs, return_tensors="pt").pixel_values
+            expected_output_image_shape = self.image_processor_tester.expected_output_image_shape(image_inputs)
+            self.assertEqual(
+                tuple(encoded_images.shape), (self.image_processor_tester.batch_size, *expected_output_image_shape)
+            )
+
+    def test_call_numpy_4_channels(self):
+        for image_processing_class in self.image_processor_list:
+            # Initialize image_processing
+            image_processor_dict = self.image_processor_dict
+            image_processor_dict["image_mean"] = [0.5, 0.5, 0.5, 0.5]
+            image_processor_dict["image_std"] = [0.5, 0.5, 0.5, 0.5]
+            image_processing = self.image_processing_class(**image_processor_dict)
+            # create random numpy tensors
+            self.image_processor_tester.num_channels = 4
+            image_inputs = self.image_processor_tester.prepare_image_inputs(equal_resolution=False, numpify=True)
+
+            for sample_images in image_inputs:
+                for image in sample_images:
+                    self.assertIsInstance(image, np.ndarray)
+
+            # Test not batched input
+            encoded_images = image_processing(
+                image_inputs[0], input_data_format="channels_last", return_tensors="pt"
+            ).pixel_values
+            expected_output_image_shape = self.image_processor_tester.expected_output_image_shape([image_inputs[0]])
+            self.assertEqual(tuple(encoded_images.shape), (1, *expected_output_image_shape))
+
+            # Test batched
+            encoded_images = image_processing(
+                image_inputs, input_data_format="channels_last", return_tensors="pt"
+            ).pixel_values
+            expected_output_image_shape = self.image_processor_tester.expected_output_image_shape(image_inputs)
+            self.assertEqual(
+                tuple(encoded_images.shape), (self.image_processor_tester.batch_size, *expected_output_image_shape)
+            )
 
     def test_call_pil(self):
-        # Initialize image_processing
-        image_processing = self.image_processing_class(**self.image_processor_dict)
-        # create random PIL images
-        image_inputs = self.image_processor_tester.prepare_image_inputs(equal_resolution=False)
-        for images in image_inputs:
-            for image in images:
-                self.assertIsInstance(image, Image.Image)
+        for image_processing_class in self.image_processor_list:
+            # Initialize image_processing
+            image_processing = self.image_processing_class(**self.image_processor_dict)
+            # create random PIL images
+            image_inputs = self.image_processor_tester.prepare_image_inputs(equal_resolution=False)
+            for images in image_inputs:
+                for image in images:
+                    self.assertIsInstance(image, Image.Image)
 
-        # Test not batched input
-        encoded_images = image_processing(image_inputs[0], return_tensors="pt").pixel_values
-        expected_output_image_shape = self.image_processor_tester.expected_output_image_shape([image_inputs[0]])
-        self.assertEqual(tuple(encoded_images.shape), (1, *expected_output_image_shape))
+            # Test not batched input
+            encoded_images = image_processing(image_inputs[0], return_tensors="pt").pixel_values
+            expected_output_image_shape = self.image_processor_tester.expected_output_image_shape([image_inputs[0]])
+            self.assertEqual(tuple(encoded_images.shape), (1, *expected_output_image_shape))
 
-        # Test batched
-        encoded_images = image_processing(image_inputs, return_tensors="pt").pixel_values
-        expected_output_image_shape = self.image_processor_tester.expected_output_image_shape(image_inputs)
-        self.assertEqual(
-            tuple(encoded_images.shape), (self.image_processor_tester.batch_size, *expected_output_image_shape)
-        )
+            # Test batched
+            encoded_images = image_processing(image_inputs, return_tensors="pt").pixel_values
+            expected_output_image_shape = self.image_processor_tester.expected_output_image_shape(image_inputs)
+            self.assertEqual(
+                tuple(encoded_images.shape), (self.image_processor_tester.batch_size, *expected_output_image_shape)
+            )
 
     def test_call_pytorch(self):
-        # Initialize image_processing
-        image_processing = self.image_processing_class(**self.image_processor_dict)
-        # create random PyTorch tensors
-        image_inputs = self.image_processor_tester.prepare_image_inputs(equal_resolution=False, torchify=True)
+        for image_processing_class in self.image_processor_list:
+            # Initialize image_processing
+            image_processing = self.image_processing_class(**self.image_processor_dict)
+            # create random PyTorch tensors
+            image_inputs = self.image_processor_tester.prepare_image_inputs(equal_resolution=False, torchify=True)
 
-        for images in image_inputs:
-            for image in images:
-                self.assertIsInstance(image, torch.Tensor)
+            for images in image_inputs:
+                for image in images:
+                    self.assertIsInstance(image, torch.Tensor)
 
-        # Test not batched input
-        encoded_images = image_processing(image_inputs[0], return_tensors="pt").pixel_values
-        expected_output_image_shape = self.image_processor_tester.expected_output_image_shape([image_inputs[0]])
-        self.assertEqual(tuple(encoded_images.shape), (1, *expected_output_image_shape))
+            # Test not batched input
+            encoded_images = image_processing(image_inputs[0], return_tensors="pt").pixel_values
+            expected_output_image_shape = self.image_processor_tester.expected_output_image_shape([image_inputs[0]])
+            self.assertEqual(tuple(encoded_images.shape), (1, *expected_output_image_shape))
 
-        # Test batched
-        expected_output_image_shape = self.image_processor_tester.expected_output_image_shape(image_inputs)
-        encoded_images = image_processing(image_inputs, return_tensors="pt").pixel_values
-        self.assertEqual(
-            tuple(encoded_images.shape),
-            (self.image_processor_tester.batch_size, *expected_output_image_shape),
-        )
+            # Test batched
+            expected_output_image_shape = self.image_processor_tester.expected_output_image_shape(image_inputs)
+            encoded_images = image_processing(image_inputs, return_tensors="pt").pixel_values
+            self.assertEqual(
+                tuple(encoded_images.shape),
+                (self.image_processor_tester.batch_size, *expected_output_image_shape),
+            )

tests/models/mask2former/test_image_processing_mask2former.py

@@ -98,6 +98,8 @@ class Mask2FormerImageProcessingTester(unittest.TestCase):
             image = image_inputs[0]
             if isinstance(image, Image.Image):
                 w, h = image.size
+            elif isinstance(image, np.ndarray):
+                h, w = image.shape[0], image.shape[1]
             else:
                 h, w = image.shape[1], image.shape[2]
             if w < h:

tests/models/maskformer/test_image_processing_maskformer.py

@@ -98,6 +98,8 @@ class MaskFormerImageProcessingTester(unittest.TestCase):
             image = image_inputs[0]
             if isinstance(image, Image.Image):
                 w, h = image.size
+            elif isinstance(image, np.ndarray):
+                h, w = image.shape[0], image.shape[1]
             else:
                 h, w = image.shape[1], image.shape[2]
             if w < h:

tests/models/oneformer/test_image_processing_oneformer.py

@@ -106,6 +106,8 @@ class OneFormerImageProcessorTester(unittest.TestCase):
             image = image_inputs[0]
             if isinstance(image, Image.Image):
                 w, h = image.size
+            elif isinstance(image, np.ndarray):
+                h, w = image.shape[0], image.shape[1]
             else:
                 h, w = image.shape[1], image.shape[2]
             if w < h:

tests/models/oneformer/test_processor_oneformer.py

@@ -143,6 +143,8 @@ class OneFormerProcessorTester(unittest.TestCase):
             image = image_inputs[0]
             if isinstance(image, Image.Image):
                 w, h = image.size
+            elif isinstance(image, np.ndarray):
+                h, w = image.shape[0], image.shape[1]
             else:
                 h, w = image.shape[1], image.shape[2]
             if w < h:

tests/models/pix2struct/test_image_processing_pix2struct.py

@@ -232,7 +232,7 @@ class Pix2StructImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase)
         for max_patch in self.image_processor_tester.max_patches:
             # Test not batched input
             encoded_images = image_processor(
-                image_inputs[0], return_tensors="pt", max_patches=max_patch, input_data_format="channels_first"
+                image_inputs[0], return_tensors="pt", max_patches=max_patch, input_data_format="channels_last"
             ).flattened_patches
             self.assertEqual(
                 encoded_images.shape,
@@ -241,7 +241,7 @@ class Pix2StructImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase)
 
             # Test batched
             encoded_images = image_processor(
-                image_inputs, return_tensors="pt", max_patches=max_patch, input_data_format="channels_first"
+                image_inputs, return_tensors="pt", max_patches=max_patch, input_data_format="channels_last"
             ).flattened_patches
             self.assertEqual(
                 encoded_images.shape,

tests/models/swin2sr/test_image_processing_swin2sr.py

@@ -72,6 +72,8 @@ class Swin2SRImageProcessingTester(unittest.TestCase):
 
         if isinstance(img, Image.Image):
             input_width, input_height = img.size
+        elif isinstance(img, np.ndarray):
+            input_height, input_width = img.shape[-3:-1]
         else:
             input_height, input_width = img.shape[-2:]
 
@@ -160,7 +162,7 @@ class Swin2SRImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
 
         # Test not batched input
         encoded_images = image_processing(
-            image_inputs[0], return_tensors="pt", input_data_format="channels_first"
+            image_inputs[0], return_tensors="pt", input_data_format="channels_last"
         ).pixel_values
         expected_output_image_shape = self.image_processor_tester.expected_output_image_shape([image_inputs[0]])
         self.assertEqual(tuple(encoded_images.shape), (1, *expected_output_image_shape))

tests/models/video_llava/test_image_processing_video_llava.py

@@ -285,7 +285,7 @@ class VideoLlavaImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase)
         encoded_images = image_processor(
             image_inputs[0],
             return_tensors="pt",
-            input_data_format="channels_first",
+            input_data_format="channels_last",
             image_mean=0,
             image_std=1,
         ).pixel_values_images
@@ -296,7 +296,7 @@ class VideoLlavaImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase)
         encoded_images = image_processor(
             image_inputs,
             return_tensors="pt",
-            input_data_format="channels_first",
+            input_data_format="channels_last",
             image_mean=0,
             image_std=1,
         ).pixel_values_images

tests/models/vilt/test_image_processing_vilt.py

@@ -16,6 +16,8 @@
 
 import unittest
 
+import numpy as np
+
 from transformers.testing_utils import require_torch, require_vision
 from transformers.utils import is_vision_available
 
@@ -78,6 +80,8 @@ class ViltImageProcessingTester(unittest.TestCase):
             image = image_inputs[0]
             if isinstance(image, Image.Image):
                 w, h = image.size
+            elif isinstance(image, np.ndarray):
+                h, w = image.shape[0], image.shape[1]
             else:
                 h, w = image.shape[1], image.shape[2]
             scale = size / min(w, h)

tests/models/vit/test_image_processing_vit.py

@@ -17,7 +17,7 @@
 import unittest
 
 from transformers.testing_utils import require_torch, require_vision
-from transformers.utils import is_vision_available
+from transformers.utils import is_torchvision_available, is_vision_available
 
 from ...test_image_processing_common import ImageProcessingTestMixin, prepare_image_inputs
 
@@ -25,6 +25,9 @@ from ...test_image_processing_common import ImageProcessingTestMixin, prepare_im
 if is_vision_available():
     from transformers import ViTImageProcessor
 
+if is_torchvision_available():
+    from transformers import ViTImageProcessorFast
+
 
 class ViTImageProcessingTester(unittest.TestCase):
     def __init__(
@@ -82,6 +85,7 @@ class ViTImageProcessingTester(unittest.TestCase):
 @require_vision
 class ViTImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
     image_processing_class = ViTImageProcessor if is_vision_available() else None
+    fast_image_processing_class = ViTImageProcessorFast if is_torchvision_available() else None
 
     def setUp(self):
         super().setUp()

tests/models/yolos/test_image_processing_yolos.py

@@ -18,6 +18,7 @@ import json
 import pathlib
 import unittest
 
+import numpy as np
 from parameterized import parameterized
 
 from transformers.testing_utils import require_torch, require_vision, slow
@@ -89,6 +90,8 @@ class YolosImageProcessingTester(unittest.TestCase):
             image = image_inputs[0]
             if isinstance(image, Image.Image):
                 width, height = image.size
+            elif isinstance(image, np.ndarray):
+                height, width = image.shape[0], image.shape[1]
             else:
                 height, width = image.shape[1], image.shape[2]