init

transformers/tests/models/sam2/test_image_processing_sam2.py

@@ -0,0 +1,243 @@
+# Copyright 2025 HuggingFace Inc.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import unittest
+
+from datasets import load_dataset
+
+from transformers.file_utils import is_torch_available, is_vision_available
+from transformers.testing_utils import require_torch, require_vision
+from transformers.utils import is_torchvision_available
+
+from ...test_image_processing_common import ImageProcessingTestMixin, prepare_image_inputs
+
+
+if is_torch_available():
+    import torch
+
+if is_vision_available() and is_torchvision_available():
+    from transformers import Sam2ImageProcessorFast
+
+
+class Sam2ImageProcessingTester:
+    def __init__(
+        self,
+        parent,
+        batch_size=7,
+        num_channels=3,
+        image_size=18,
+        min_resolution=30,
+        max_resolution=400,
+        mask_size=None,
+        do_resize=True,
+        size=None,
+        do_normalize=True,
+        image_mean=[0.5, 0.5, 0.5],
+        image_std=[0.5, 0.5, 0.5],
+    ):
+        size = size if size is not None else {"height": 20, "width": 20}
+        mask_size = mask_size if mask_size is not None else {"height": 12, "width": 12}
+        self.parent = parent
+        self.batch_size = batch_size
+        self.num_channels = num_channels
+        self.image_size = image_size
+        self.min_resolution = min_resolution
+        self.max_resolution = max_resolution
+        self.mask_size = mask_size
+        self.do_resize = do_resize
+        self.size = size
+        self.do_normalize = do_normalize
+        self.image_mean = image_mean
+        self.image_std = image_std
+
+    def prepare_image_processor_dict(self):
+        return {
+            "image_mean": self.image_mean,
+            "image_std": self.image_std,
+            "do_normalize": self.do_normalize,
+            "do_resize": self.do_resize,
+            "size": self.size,
+            "mask_size": self.mask_size,
+        }
+
+    def expected_output_image_shape(self, images):
+        return self.num_channels, self.size["height"], self.size["width"]
+
+    def prepare_image_inputs(self, equal_resolution=False, numpify=False, torchify=False):
+        return prepare_image_inputs(
+            batch_size=self.batch_size,
+            num_channels=self.num_channels,
+            min_resolution=self.min_resolution,
+            max_resolution=self.max_resolution,
+            equal_resolution=equal_resolution,
+            numpify=numpify,
+            torchify=torchify,
+        )
+
+
+# Copied from transformers.tests.models.beit.test_image_processing_beit.prepare_semantic_single_inputs
+def prepare_semantic_single_inputs():
+    ds = load_dataset("hf-internal-testing/fixtures_ade20k", split="test")
+    example = ds[0]
+    return example["image"], example["map"]
+
+
+# Copied from transformers.tests.models.beit.test_image_processing_beit.prepare_semantic_batch_inputs
+def prepare_semantic_batch_inputs():
+    ds = load_dataset("hf-internal-testing/fixtures_ade20k", split="test")
+    return list(ds["image"][:2]), list(ds["map"][:2])
+
+
+@require_torch
+@require_vision
+class SamImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
+    fast_image_processing_class = Sam2ImageProcessorFast if is_torchvision_available() else None
+    test_slow_image_processor = False
+
+    def setUp(self):
+        super().setUp()
+        self.image_processor_tester = Sam2ImageProcessingTester(self)
+
+    @property
+    def image_processor_dict(self):
+        return self.image_processor_tester.prepare_image_processor_dict()
+
+    def test_image_processor_properties(self):
+        for image_processing_class in self.image_processor_list:
+            image_processing = image_processing_class(**self.image_processor_dict)
+            self.assertTrue(hasattr(image_processing, "image_mean"))
+            self.assertTrue(hasattr(image_processing, "image_std"))
+            self.assertTrue(hasattr(image_processing, "do_normalize"))
+            self.assertTrue(hasattr(image_processing, "do_resize"))
+            self.assertTrue(hasattr(image_processing, "size"))
+            self.assertTrue(hasattr(image_processing, "do_rescale"))
+            self.assertTrue(hasattr(image_processing, "rescale_factor"))
+            self.assertTrue(hasattr(image_processing, "mask_size"))
+
+    def test_image_processor_from_dict_with_kwargs(self):
+        for image_processing_class in self.image_processor_list:
+            image_processing_class = image_processing_class(**self.image_processor_dict)
+            image_processor = image_processing_class.from_dict(self.image_processor_dict)
+            self.assertEqual(image_processor.size, {"height": 20, "width": 20})
+
+            image_processor = image_processing_class.from_dict(self.image_processor_dict, size=42)
+            self.assertEqual(image_processor.size, {"height": 42, "width": 42})
+
+    def test_call_segmentation_maps(self):
+        for image_processing_class in self.image_processor_list:
+            # Initialize image_processor
+            image_processor = 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)
+            maps = []
+            for image in image_inputs:
+                self.assertIsInstance(image, torch.Tensor)
+                maps.append(torch.zeros(image.shape[-2:]).long())
+
+            # Test not batched input
+            encoding = image_processor(image_inputs[0], maps[0], return_tensors="pt")
+            self.assertEqual(
+                encoding["pixel_values"].shape,
+                (
+                    1,
+                    self.image_processor_tester.num_channels,
+                    self.image_processor_tester.size["height"],
+                    self.image_processor_tester.size["width"],
+                ),
+            )
+            self.assertEqual(
+                encoding["labels"].shape,
+                (
+                    1,
+                    self.image_processor_tester.mask_size["height"],
+                    self.image_processor_tester.mask_size["width"],
+                ),
+            )
+            self.assertEqual(encoding["labels"].dtype, torch.long)
+            self.assertTrue(encoding["labels"].min().item() >= 0)
+            self.assertTrue(encoding["labels"].max().item() <= 255)
+
+            # Test batched
+            encoding = image_processor(image_inputs, maps, return_tensors="pt")
+            self.assertEqual(
+                encoding["pixel_values"].shape,
+                (
+                    self.image_processor_tester.batch_size,
+                    self.image_processor_tester.num_channels,
+                    self.image_processor_tester.size["height"],
+                    self.image_processor_tester.size["width"],
+                ),
+            )
+            self.assertEqual(
+                encoding["labels"].shape,
+                (
+                    self.image_processor_tester.batch_size,
+                    self.image_processor_tester.mask_size["height"],
+                    self.image_processor_tester.mask_size["width"],
+                ),
+            )
+            self.assertEqual(encoding["labels"].dtype, torch.long)
+            self.assertTrue(encoding["labels"].min().item() >= 0)
+            self.assertTrue(encoding["labels"].max().item() <= 255)
+
+            # Test not batched input (PIL images)
+            image, segmentation_map = prepare_semantic_single_inputs()
+
+            encoding = image_processor(image, segmentation_map, return_tensors="pt")
+            self.assertEqual(
+                encoding["pixel_values"].shape,
+                (
+                    1,
+                    self.image_processor_tester.num_channels,
+                    self.image_processor_tester.size["height"],
+                    self.image_processor_tester.size["width"],
+                ),
+            )
+            self.assertEqual(
+                encoding["labels"].shape,
+                (
+                    1,
+                    self.image_processor_tester.mask_size["height"],
+                    self.image_processor_tester.mask_size["width"],
+                ),
+            )
+            self.assertEqual(encoding["labels"].dtype, torch.long)
+            self.assertTrue(encoding["labels"].min().item() >= 0)
+            self.assertTrue(encoding["labels"].max().item() <= 255)
+
+            # Test batched input (PIL images)
+            images, segmentation_maps = prepare_semantic_batch_inputs()
+
+            encoding = image_processor(images, segmentation_maps, return_tensors="pt")
+            self.assertEqual(
+                encoding["pixel_values"].shape,
+                (
+                    2,
+                    self.image_processor_tester.num_channels,
+                    self.image_processor_tester.size["height"],
+                    self.image_processor_tester.size["width"],
+                ),
+            )
+            self.assertEqual(
+                encoding["labels"].shape,
+                (
+                    2,
+                    self.image_processor_tester.mask_size["height"],
+                    self.image_processor_tester.mask_size["width"],
+                ),
+            )
+            self.assertEqual(encoding["labels"].dtype, torch.long)
+            self.assertTrue(encoding["labels"].min().item() >= 0)
+            self.assertTrue(encoding["labels"].max().item() <= 255)

transformers/tests/models/sam2/test_processor_sam2.py

@@ -0,0 +1,144 @@
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import shutil
+import tempfile
+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_vision_available
+
+
+if is_vision_available():
+    from transformers import AutoProcessor, Sam2ImageProcessorFast, Sam2Processor
+
+if is_torch_available():
+    import torch
+
+
+@require_vision
+@require_torchvision
+class Sam2ProcessorTest(unittest.TestCase):
+    def setUp(self):
+        self.tmpdirname = tempfile.mkdtemp()
+        image_processor = Sam2ImageProcessorFast()
+        processor = Sam2Processor(image_processor)
+        processor.save_pretrained(self.tmpdirname)
+
+    def get_image_processor(self, **kwargs):
+        return AutoProcessor.from_pretrained(self.tmpdirname, **kwargs).image_processor
+
+    def tearDown(self):
+        shutil.rmtree(self.tmpdirname)
+
+    def prepare_image_inputs(self):
+        """This function prepares a list of PIL images, or a list of numpy arrays if one specifies numpify=True,
+        or a list of PyTorch tensors if one specifies torchify=True.
+        """
+        image_inputs = torch.randint(0, 256, size=(1, 3, 30, 400), dtype=torch.uint8)
+        # image_inputs = [Image.fromarray(np.moveaxis(x, 0, -1)) for x in image_inputs]
+        return image_inputs
+
+    def prepare_mask_inputs(self):
+        """This function prepares a list of PIL images, or a list of numpy arrays if one specifies numpify=True,
+        or a list of PyTorch tensors if one specifies torchify=True.
+        """
+        mask_inputs = torch.randint(0, 256, size=(1, 30, 400), dtype=torch.uint8)
+        # mask_inputs = [Image.fromarray(x) for x in mask_inputs]
+        return mask_inputs
+
+    def test_save_load_pretrained_additional_features(self):
+        image_processor = self.get_image_processor()
+
+        processor = Sam2Processor(image_processor=image_processor)
+        processor.save_pretrained(self.tmpdirname)
+
+        image_processor_add_kwargs = self.get_image_processor(do_normalize=False, padding_value=1.0)
+
+        processor = Sam2Processor.from_pretrained(self.tmpdirname, do_normalize=False, padding_value=1.0)
+
+        self.assertEqual(processor.image_processor.to_json_string(), image_processor_add_kwargs.to_json_string())
+        self.assertIsInstance(processor.image_processor, Sam2ImageProcessorFast)
+
+    def test_image_processor_no_masks(self):
+        image_processor = self.get_image_processor()
+
+        processor = Sam2Processor(image_processor=image_processor)
+
+        image_input = self.prepare_image_inputs()
+
+        input_feat_extract = image_processor(image_input)
+        input_processor = processor(images=image_input)
+
+        for key in input_feat_extract.keys():
+            if key == "pixel_values":
+                for input_feat_extract_item, input_processor_item in zip(
+                    input_feat_extract[key], input_processor[key]
+                ):
+                    np.testing.assert_array_equal(input_feat_extract_item, input_processor_item)
+            else:
+                self.assertEqual(input_feat_extract[key], input_processor[key])
+
+        for image in input_feat_extract.pixel_values:
+            self.assertEqual(image.shape, (3, 1024, 1024))
+
+        for original_size in input_feat_extract.original_sizes:
+            np.testing.assert_array_equal(original_size, np.array([30, 400]))
+
+    def test_image_processor_with_masks(self):
+        image_processor = self.get_image_processor()
+
+        processor = Sam2Processor(image_processor=image_processor)
+
+        image_input = self.prepare_image_inputs()
+        mask_input = self.prepare_mask_inputs()
+
+        input_feat_extract = image_processor(images=image_input, segmentation_maps=mask_input, return_tensors="pt")
+        input_processor = processor(images=image_input, segmentation_maps=mask_input, return_tensors="pt")
+
+        for key in input_feat_extract.keys():
+            self.assertAlmostEqual(input_feat_extract[key].sum(), input_processor[key].sum(), delta=1e-2)
+
+        for label in input_feat_extract.labels:
+            self.assertEqual(label.shape, (256, 256))
+
+    @require_torch
+    def test_post_process_masks(self):
+        image_processor = self.get_image_processor()
+
+        processor = Sam2Processor(image_processor=image_processor)
+        dummy_masks = [torch.ones((1, 3, 5, 5))]
+
+        original_sizes = [[1764, 2646]]
+
+        masks = processor.post_process_masks(dummy_masks, original_sizes)
+        self.assertEqual(masks[0].shape, (1, 3, 1764, 2646))
+
+        masks = processor.post_process_masks(dummy_masks, torch.tensor(original_sizes))
+        self.assertEqual(masks[0].shape, (1, 3, 1764, 2646))
+
+        # should also work with np
+        dummy_masks = [np.ones((1, 3, 5, 5))]
+        masks = processor.post_process_masks(dummy_masks, np.array(original_sizes))
+
+        self.assertEqual(masks[0].shape, (1, 3, 1764, 2646))
+
+        dummy_masks = [[1, 0], [0, 1]]
+        with self.assertRaises(ValueError):
+            masks = processor.post_process_masks(dummy_masks, np.array(original_sizes))