[Fast Processor] BEiT (#37005)
* adding fast processor for beit * adding resample * address review issues and add segmentation maps logic * style * chore: adding tests * reduce label test * adding batched tests * Update src/transformers/models/beit/image_processing_beit_fast.py Co-authored-by: Yoni Gozlan <74535834+yonigozlan@users.noreply.github.com> * fix imports and make segmentation masks * fix tests * build segmentation maps * all tests pass * style * style fix * style * chore: delete demo.py file * review suggestions * Update docs/source/en/model_doc/beit.md Co-authored-by: Yoni Gozlan <74535834+yonigozlan@users.noreply.github.com> --------- Co-authored-by: Yoni Gozlan <74535834+yonigozlan@users.noreply.github.com>
This commit is contained in:
Aritra Roy Gosthipaty
committed by
GitHub
GitHub
parent
ebbe9b12dd
commit
3c0796aaea
@@ -18,7 +18,7 @@ import unittest
|
||||
from datasets import load_dataset
|
||||
|
||||
from transformers.testing_utils import require_torch, require_vision
|
||||
from transformers.utils import is_torch_available, is_vision_available
|
||||
from transformers.utils import is_torch_available, is_torchvision_available, is_vision_available
|
||||
|
||||
from ...test_image_processing_common import ImageProcessingTestMixin, prepare_image_inputs
|
||||
|
||||
@@ -31,6 +31,9 @@ if is_vision_available():
|
||||
|
||||
from transformers import BeitImageProcessor
|
||||
|
||||
if is_torchvision_available():
|
||||
from transformers import BeitImageProcessorFast
|
||||
|
||||
|
||||
class BeitImageProcessingTester:
|
||||
def __init__(
|
||||
@@ -118,6 +121,7 @@ def prepare_semantic_batch_inputs():
|
||||
@require_vision
|
||||
class BeitImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
|
||||
image_processing_class = BeitImageProcessor if is_vision_available() else None
|
||||
fast_image_processing_class = BeitImageProcessorFast if is_torchvision_available() else None
|
||||
|
||||
def setUp(self):
|
||||
super().setUp()
|
||||
@@ -128,159 +132,196 @@ class BeitImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
|
||||
return self.image_processor_tester.prepare_image_processor_dict()
|
||||
|
||||
def test_image_processor_properties(self):
|
||||
image_processing = self.image_processing_class(**self.image_processor_dict)
|
||||
self.assertTrue(hasattr(image_processing, "do_resize"))
|
||||
self.assertTrue(hasattr(image_processing, "size"))
|
||||
self.assertTrue(hasattr(image_processing, "do_center_crop"))
|
||||
self.assertTrue(hasattr(image_processing, "center_crop"))
|
||||
self.assertTrue(hasattr(image_processing, "do_normalize"))
|
||||
self.assertTrue(hasattr(image_processing, "image_mean"))
|
||||
self.assertTrue(hasattr(image_processing, "image_std"))
|
||||
self.assertTrue(hasattr(image_processing, "do_reduce_labels"))
|
||||
for image_processing_class in self.image_processor_list:
|
||||
image_processing = image_processing_class(**self.image_processor_dict)
|
||||
self.assertTrue(hasattr(image_processing, "do_resize"))
|
||||
self.assertTrue(hasattr(image_processing, "size"))
|
||||
self.assertTrue(hasattr(image_processing, "do_center_crop"))
|
||||
self.assertTrue(hasattr(image_processing, "center_crop"))
|
||||
self.assertTrue(hasattr(image_processing, "do_normalize"))
|
||||
self.assertTrue(hasattr(image_processing, "image_mean"))
|
||||
self.assertTrue(hasattr(image_processing, "image_std"))
|
||||
self.assertTrue(hasattr(image_processing, "do_reduce_labels"))
|
||||
|
||||
def test_image_processor_from_dict_with_kwargs(self):
|
||||
image_processor = self.image_processing_class.from_dict(self.image_processor_dict)
|
||||
self.assertEqual(image_processor.size, {"height": 20, "width": 20})
|
||||
self.assertEqual(image_processor.crop_size, {"height": 18, "width": 18})
|
||||
self.assertEqual(image_processor.do_reduce_labels, False)
|
||||
for image_processing_class in self.image_processor_list:
|
||||
image_processor = image_processing_class.from_dict(self.image_processor_dict)
|
||||
self.assertEqual(image_processor.size, {"height": 20, "width": 20})
|
||||
self.assertEqual(image_processor.crop_size, {"height": 18, "width": 18})
|
||||
self.assertEqual(image_processor.do_reduce_labels, False)
|
||||
|
||||
image_processor = self.image_processing_class.from_dict(
|
||||
self.image_processor_dict, size=42, crop_size=84, do_reduce_labels=True
|
||||
)
|
||||
self.assertEqual(image_processor.size, {"height": 42, "width": 42})
|
||||
self.assertEqual(image_processor.crop_size, {"height": 84, "width": 84})
|
||||
self.assertEqual(image_processor.do_reduce_labels, True)
|
||||
image_processor = image_processing_class.from_dict(
|
||||
self.image_processor_dict, size=42, crop_size=84, do_reduce_labels=True
|
||||
)
|
||||
self.assertEqual(image_processor.size, {"height": 42, "width": 42})
|
||||
self.assertEqual(image_processor.crop_size, {"height": 84, "width": 84})
|
||||
self.assertEqual(image_processor.do_reduce_labels, True)
|
||||
|
||||
def test_call_segmentation_maps(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)
|
||||
maps = []
|
||||
for image in image_inputs:
|
||||
self.assertIsInstance(image, torch.Tensor)
|
||||
maps.append(torch.zeros(image.shape[-2:]).long())
|
||||
for image_processing_class in self.image_processor_list:
|
||||
# Initialize image_processing
|
||||
image_processing = 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_processing(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.crop_size["height"],
|
||||
self.image_processor_tester.crop_size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(
|
||||
encoding["labels"].shape,
|
||||
(
|
||||
1,
|
||||
self.image_processor_tester.crop_size["height"],
|
||||
self.image_processor_tester.crop_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
|
||||
encoding = image_processing(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.crop_size["height"],
|
||||
self.image_processor_tester.crop_size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(
|
||||
encoding["labels"].shape,
|
||||
(
|
||||
1,
|
||||
self.image_processor_tester.crop_size["height"],
|
||||
self.image_processor_tester.crop_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_processing(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.crop_size["height"],
|
||||
self.image_processor_tester.crop_size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(
|
||||
encoding["labels"].shape,
|
||||
(
|
||||
self.image_processor_tester.batch_size,
|
||||
self.image_processor_tester.crop_size["height"],
|
||||
self.image_processor_tester.crop_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_processing(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.crop_size["height"],
|
||||
self.image_processor_tester.crop_size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(
|
||||
encoding["labels"].shape,
|
||||
(
|
||||
self.image_processor_tester.batch_size,
|
||||
self.image_processor_tester.crop_size["height"],
|
||||
self.image_processor_tester.crop_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()
|
||||
# Test not batched input (PIL images)
|
||||
image, segmentation_map = prepare_semantic_single_inputs()
|
||||
|
||||
encoding = image_processing(image, segmentation_map, return_tensors="pt")
|
||||
self.assertEqual(
|
||||
encoding["pixel_values"].shape,
|
||||
(
|
||||
1,
|
||||
self.image_processor_tester.num_channels,
|
||||
self.image_processor_tester.crop_size["height"],
|
||||
self.image_processor_tester.crop_size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(
|
||||
encoding["labels"].shape,
|
||||
(
|
||||
1,
|
||||
self.image_processor_tester.crop_size["height"],
|
||||
self.image_processor_tester.crop_size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(encoding["labels"].dtype, torch.long)
|
||||
self.assertTrue(encoding["labels"].min().item() >= 0)
|
||||
self.assertTrue(encoding["labels"].max().item() <= 255)
|
||||
encoding = image_processing(image, segmentation_map, return_tensors="pt")
|
||||
self.assertEqual(
|
||||
encoding["pixel_values"].shape,
|
||||
(
|
||||
1,
|
||||
self.image_processor_tester.num_channels,
|
||||
self.image_processor_tester.crop_size["height"],
|
||||
self.image_processor_tester.crop_size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(
|
||||
encoding["labels"].shape,
|
||||
(
|
||||
1,
|
||||
self.image_processor_tester.crop_size["height"],
|
||||
self.image_processor_tester.crop_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()
|
||||
# Test batched input (PIL images)
|
||||
images, segmentation_maps = prepare_semantic_batch_inputs()
|
||||
|
||||
encoding = image_processing(images, segmentation_maps, return_tensors="pt")
|
||||
self.assertEqual(
|
||||
encoding["pixel_values"].shape,
|
||||
(
|
||||
2,
|
||||
self.image_processor_tester.num_channels,
|
||||
self.image_processor_tester.crop_size["height"],
|
||||
self.image_processor_tester.crop_size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(
|
||||
encoding["labels"].shape,
|
||||
(
|
||||
2,
|
||||
self.image_processor_tester.crop_size["height"],
|
||||
self.image_processor_tester.crop_size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(encoding["labels"].dtype, torch.long)
|
||||
self.assertTrue(encoding["labels"].min().item() >= 0)
|
||||
self.assertTrue(encoding["labels"].max().item() <= 255)
|
||||
encoding = image_processing(images, segmentation_maps, return_tensors="pt")
|
||||
self.assertEqual(
|
||||
encoding["pixel_values"].shape,
|
||||
(
|
||||
2,
|
||||
self.image_processor_tester.num_channels,
|
||||
self.image_processor_tester.crop_size["height"],
|
||||
self.image_processor_tester.crop_size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(
|
||||
encoding["labels"].shape,
|
||||
(
|
||||
2,
|
||||
self.image_processor_tester.crop_size["height"],
|
||||
self.image_processor_tester.crop_size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(encoding["labels"].dtype, torch.long)
|
||||
self.assertTrue(encoding["labels"].min().item() >= 0)
|
||||
self.assertTrue(encoding["labels"].max().item() <= 255)
|
||||
|
||||
def test_reduce_labels(self):
|
||||
# Initialize image_processing
|
||||
image_processing = self.image_processing_class(**self.image_processor_dict)
|
||||
for image_processing_class in self.image_processor_list:
|
||||
# Initialize image_processing
|
||||
image_processing = image_processing_class(**self.image_processor_dict)
|
||||
|
||||
# ADE20k has 150 classes, and the background is included, so labels should be between 0 and 150
|
||||
image, map = prepare_semantic_single_inputs()
|
||||
encoding = image_processing(image, map, return_tensors="pt")
|
||||
self.assertTrue(encoding["labels"].min().item() >= 0)
|
||||
self.assertTrue(encoding["labels"].max().item() <= 150)
|
||||
# ADE20k has 150 classes, and the background is included, so labels should be between 0 and 150
|
||||
image, map = prepare_semantic_single_inputs()
|
||||
encoding = image_processing(image, map, return_tensors="pt")
|
||||
self.assertTrue(encoding["labels"].min().item() >= 0)
|
||||
self.assertTrue(encoding["labels"].max().item() <= 150)
|
||||
|
||||
image_processing.do_reduce_labels = True
|
||||
encoding = image_processing(image, map, return_tensors="pt")
|
||||
self.assertTrue(encoding["labels"].min().item() >= 0)
|
||||
self.assertTrue(encoding["labels"].max().item() <= 255)
|
||||
image_processing.do_reduce_labels = True
|
||||
encoding = image_processing(image, map, return_tensors="pt")
|
||||
self.assertTrue(encoding["labels"].min().item() >= 0)
|
||||
self.assertTrue(encoding["labels"].max().item() <= 255)
|
||||
|
||||
def test_removed_deprecated_kwargs(self):
|
||||
image_processor_dict = dict(self.image_processor_dict)
|
||||
image_processor_dict.pop("do_reduce_labels", None)
|
||||
image_processor_dict["reduce_labels"] = True
|
||||
def test_slow_fast_equivalence(self):
|
||||
if not self.test_slow_image_processor or not self.test_fast_image_processor:
|
||||
self.skipTest(reason="Skipping slow/fast equivalence test")
|
||||
|
||||
# test we are able to create the image processor with the deprecated kwargs
|
||||
image_processor = self.image_processing_class(**image_processor_dict)
|
||||
self.assertEqual(image_processor.do_reduce_labels, True)
|
||||
if self.image_processing_class is None or self.fast_image_processing_class is None:
|
||||
self.skipTest(reason="Skipping slow/fast equivalence test as one of the image processors is not defined")
|
||||
|
||||
# test we still support reduce_labels with config
|
||||
image_processor = self.image_processing_class.from_dict(image_processor_dict)
|
||||
self.assertEqual(image_processor.do_reduce_labels, True)
|
||||
dummy_image, dummy_map = prepare_semantic_single_inputs()
|
||||
|
||||
image_processor_slow = self.image_processing_class(**self.image_processor_dict)
|
||||
image_processor_fast = self.fast_image_processing_class(**self.image_processor_dict)
|
||||
|
||||
image_encoding_slow = image_processor_slow(dummy_image, segmentation_maps=dummy_map, return_tensors="pt")
|
||||
image_encoding_fast = image_processor_fast(dummy_image, segmentation_maps=dummy_map, return_tensors="pt")
|
||||
|
||||
self.assertTrue(torch.allclose(image_encoding_slow.pixel_values, image_encoding_fast.pixel_values, atol=1e-1))
|
||||
self.assertLessEqual(
|
||||
torch.mean(torch.abs(image_encoding_slow.pixel_values - image_encoding_fast.pixel_values)).item(), 1e-3
|
||||
)
|
||||
self.assertTrue(torch.allclose(image_encoding_slow.labels, image_encoding_fast.labels, atol=1e-1))
|
||||
|
||||
def test_slow_fast_equivalence_batched(self):
|
||||
if not self.test_slow_image_processor or not self.test_fast_image_processor:
|
||||
self.skipTest(reason="Skipping slow/fast equivalence test")
|
||||
|
||||
if self.image_processing_class is None or self.fast_image_processing_class is None:
|
||||
self.skipTest(reason="Skipping slow/fast equivalence test as one of the image processors is not defined")
|
||||
|
||||
if hasattr(self.image_processor_tester, "do_center_crop") and self.image_processor_tester.do_center_crop:
|
||||
self.skipTest(
|
||||
reason="Skipping as do_center_crop is True and center_crop functions are not equivalent for fast and slow processors"
|
||||
)
|
||||
|
||||
dummy_images, dummy_maps = prepare_semantic_batch_inputs()
|
||||
|
||||
image_processor_slow = self.image_processing_class(**self.image_processor_dict)
|
||||
image_processor_fast = self.fast_image_processing_class(**self.image_processor_dict)
|
||||
|
||||
encoding_slow = image_processor_slow(dummy_images, segmentation_maps=dummy_maps, return_tensors="pt")
|
||||
encoding_fast = image_processor_fast(dummy_images, segmentation_maps=dummy_maps, return_tensors="pt")
|
||||
|
||||
self.assertTrue(torch.allclose(encoding_slow.pixel_values, encoding_fast.pixel_values, atol=1e-1))
|
||||
self.assertLessEqual(
|
||||
torch.mean(torch.abs(encoding_slow.pixel_values - encoding_fast.pixel_values)).item(), 1e-3
|
||||
)
|
||||
|
||||
Reference in New Issue
Block a user