SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Add DetrImageProcessorFast (#34063)

Browse Source 
* add fully functionning image_processing_detr_fast

* Create tensors on the correct device

* fix copies

* fix doc

* add tests equivalence cpu gpu

* fix doc en

* add relative imports and copied from

* Fix copies and nit
This commit is contained in:
Yoni Gozlan
2024-10-21 09:05:05 -04:00
committed by GitHub
parent 24bdc94da5
commit a4122813d1
16 changed files with 2779 additions and 1047 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
docs/source/en/model_doc/detr.md
View File

@@ -181,6 +181,15 @@ If you're interested in submitting a resource to be included here, please feel f
- post_process_instance_segmentation
- post_process_panoptic_segmentation
## DetrImageProcessorFast
[[autodoc]] DetrImageProcessorFast
- preprocess
- post_process_object_detection
- post_process_semantic_segmentation
- post_process_instance_segmentation
- post_process_panoptic_segmentation
## DetrFeatureExtractor
[[autodoc]] DetrFeatureExtractor

9
docs/source/ja/model_doc/detr.md
View File

@@ -184,6 +184,15 @@ DETR の使用を開始するのに役立つ公式 Hugging Face およびコミ
- post_process_instance_segmentation
- post_process_panoptic_segmentation
## DetrImageProcessorFast
[[autodoc]] DetrImageProcessorFast
- preprocess
- post_process_object_detection
- post_process_semantic_segmentation
- post_process_instance_segmentation
- post_process_panoptic_segmentation
## DetrFeatureExtractor
[[autodoc]] DetrFeatureExtractor

4
src/transformers/__init__.py
View File

@@ -1191,7 +1191,7 @@ else:
_import_structure["models.deprecated.efficientformer"].append("EfficientFormerImageProcessor")
_import_structure["models.deprecated.tvlt"].append("TvltImageProcessor")
_import_structure["models.deprecated.vit_hybrid"].extend(["ViTHybridImageProcessor"])
_import_structure["models.detr"].extend(["DetrFeatureExtractor", "DetrImageProcessor"])
_import_structure["models.detr"].extend(["DetrFeatureExtractor", "DetrImageProcessor", "DetrImageProcessorFast"])
_import_structure["models.donut"].extend(["DonutFeatureExtractor", "DonutImageProcessor"])
_import_structure["models.dpt"].extend(["DPTFeatureExtractor", "DPTImageProcessor"])
_import_structure["models.efficientnet"].append("EfficientNetImageProcessor")
@@ -6090,7 +6090,7 @@ if TYPE_CHECKING:
from .models.deprecated.efficientformer import EfficientFormerImageProcessor
from .models.deprecated.tvlt import TvltImageProcessor
from .models.deprecated.vit_hybrid import ViTHybridImageProcessor
from .models.detr import DetrFeatureExtractor, DetrImageProcessor
from .models.detr import DetrFeatureExtractor, DetrImageProcessor, DetrImageProcessorFast
from .models.donut import DonutFeatureExtractor, DonutImageProcessor
from .models.dpt import DPTFeatureExtractor, DPTImageProcessor
from .models.efficientnet import EfficientNetImageProcessor

5
src/transformers/image_transforms.py
View File

@@ -32,6 +32,7 @@ from .utils.import_utils import (
is_tf_available,
is_torch_available,
is_torchvision_available,
is_torchvision_v2_available,
is_vision_available,
requires_backends,
)
@@ -51,7 +52,9 @@ if is_tf_available():
if is_flax_available():
import jax.numpy as jnp
if is_torchvision_available():
if is_torchvision_v2_available():
from torchvision.transforms.v2 import functional as F
elif is_torchvision_available():
from torchvision.transforms import functional as F

2
src/transformers/models/auto/image_processing_auto.py
View File

@@ -72,7 +72,7 @@ else:
("deit", ("DeiTImageProcessor",)),
("depth_anything", ("DPTImageProcessor",)),
("deta", ("DetaImageProcessor",)),
("detr", ("DetrImageProcessor",)),
("detr", ("DetrImageProcessor", "DetrImageProcessorFast")),
("dinat", ("ViTImageProcessor", "ViTImageProcessorFast")),
("dinov2", ("BitImageProcessor",)),
("donut-swin", ("DonutImageProcessor",)),

2
src/transformers/models/detr/__init__.py
View File

@@ -27,6 +27,7 @@ except OptionalDependencyNotAvailable:
else:
_import_structure["feature_extraction_detr"] = ["DetrFeatureExtractor"]
_import_structure["image_processing_detr"] = ["DetrImageProcessor"]
_import_structure["image_processing_detr_fast"] = ["DetrImageProcessorFast"]
try:
if not is_torch_available():
@@ -53,6 +54,7 @@ if TYPE_CHECKING:
else:
from .feature_extraction_detr import DetrFeatureExtractor
from .image_processing_detr import DetrImageProcessor
from .image_processing_detr_fast import DetrImageProcessorFast
try:
if not is_torch_available():

1546
src/transformers/models/detr/image_processing_detr_fast.py Normal file
View File

File diff suppressed because it is too large Load Diff

1
src/transformers/utils/__init__.py
View File

@@ -225,6 +225,7 @@ from .import_utils import (
is_torchdynamo_available,
is_torchdynamo_compiling,
is_torchvision_available,
is_torchvision_v2_available,
is_training_run_on_sagemaker,
is_uroman_available,
is_vision_available,

7
src/transformers/utils/dummy_vision_objects.py
View File

@@ -191,6 +191,13 @@ class DetrImageProcessor(metaclass=DummyObject):
requires_backends(self, ["vision"])
class DetrImageProcessorFast(metaclass=DummyObject):
_backends = ["vision"]
def __init__(self, *args, **kwargs):
requires_backends(self, ["vision"])
class DonutFeatureExtractor(metaclass=DummyObject):
_backends = ["vision"]

10
src/transformers/utils/import_utils.py
View File

@@ -186,7 +186,7 @@ _tokenizers_available = _is_package_available("tokenizers")
_torchaudio_available = _is_package_available("torchaudio")
_torchao_available = _is_package_available("torchao")
_torchdistx_available = _is_package_available("torchdistx")
_torchvision_available = _is_package_available("torchvision")
_torchvision_available, _torchvision_version = _is_package_available("torchvision", return_version=True)
_mlx_available = _is_package_available("mlx")
_hqq_available, _hqq_version = _is_package_available("hqq", return_version=True)
_tiktoken_available = _is_package_available("tiktoken")
@@ -362,6 +362,14 @@ def is_torchvision_available():
return _torchvision_available
def is_torchvision_v2_available():
if not is_torchvision_available():
return False
# NOTE: We require torchvision>=0.15 as v2 transforms are available from this version: https://pytorch.org/vision/stable/transforms.html#v1-or-v2-which-one-should-i-use
return version.parse(_torchvision_version) >= version.parse("0.15")
def is_galore_torch_available():
return _galore_torch_available

17
tests/models/conditional_detr/test_image_processing_conditional_detr.py
View File

@@ -282,7 +282,8 @@ class ConditionalDetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcess
images = [image_0, image_1]
annotations = [annotations_0, annotations_1]
image_processing = ConditionalDetrImageProcessor()
for image_processing_class in self.image_processor_list:
image_processing = image_processing_class()
encoding = image_processing(
images=images,
annotations=annotations,
@@ -402,8 +403,9 @@ class ConditionalDetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcess
images = [image_0, image_1]
annotations = [annotation_0, annotation_1]
for image_processing_class in self.image_processor_list:
# encode them
image_processing = ConditionalDetrImageProcessor(format="coco_panoptic")
image_processing = image_processing_class(format="coco_panoptic")
encoding = image_processing(
images=images,
annotations=annotations,
@@ -498,10 +500,11 @@ class ConditionalDetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcess
# Copied from tests.models.detr.test_image_processing_detr.DetrImageProcessingTest.test_max_width_max_height_resizing_and_pad_strategy with Detr->ConditionalDetr
def test_max_width_max_height_resizing_and_pad_strategy(self):
for image_processing_class in self.image_processor_list:
image_1 = torch.ones([200, 100, 3], dtype=torch.uint8)
# do_pad=False, max_height=100, max_width=100, image=200x100 -> 100x50
image_processor = ConditionalDetrImageProcessor(
image_processor = image_processing_class(
size={"max_height": 100, "max_width": 100},
do_pad=False,
)
@@ -509,21 +512,21 @@ class ConditionalDetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcess
self.assertEqual(inputs["pixel_values"].shape, torch.Size([1, 3, 100, 50]))
# do_pad=False, max_height=300, max_width=100, image=200x100 -> 200x100
image_processor = ConditionalDetrImageProcessor(
image_processor = image_processing_class(
size={"max_height": 300, "max_width": 100},
do_pad=False,
)
inputs = image_processor(images=[image_1], return_tensors="pt")
# do_pad=True, max_height=100, max_width=100, image=200x100 -> 100x100
image_processor = ConditionalDetrImageProcessor(
image_processor = image_processing_class(
size={"max_height": 100, "max_width": 100}, do_pad=True, pad_size={"height": 100, "width": 100}
)
inputs = image_processor(images=[image_1], return_tensors="pt")
self.assertEqual(inputs["pixel_values"].shape, torch.Size([1, 3, 100, 100]))
# do_pad=True, max_height=300, max_width=100, image=200x100 -> 300x100
image_processor = ConditionalDetrImageProcessor(
image_processor = image_processing_class(
size={"max_height": 300, "max_width": 100},
do_pad=True,
pad_size={"height": 301, "width": 101},
@@ -535,7 +538,7 @@ class ConditionalDetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcess
image_2 = torch.ones([100, 150, 3], dtype=torch.uint8)
# do_pad=True, max_height=150, max_width=100, images=[200x100, 100x150] -> 150x100
image_processor = ConditionalDetrImageProcessor(
image_processor = image_processing_class(
size={"max_height": 150, "max_width": 100},
do_pad=True,
pad_size={"height": 150, "width": 100},

17
tests/models/deformable_detr/test_image_processing_deformable_detr.py
View File

@@ -284,7 +284,8 @@ class DeformableDetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessi
images = [image_0, image_1]
annotations = [annotations_0, annotations_1]
image_processing = DeformableDetrImageProcessor()
for image_processing_class in self.image_processor_list:
image_processing = image_processing_class()
encoding = image_processing(
images=images,
annotations=annotations,
@@ -404,8 +405,9 @@ class DeformableDetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessi
images = [image_0, image_1]
annotations = [annotation_0, annotation_1]
for image_processing_class in self.image_processor_list:
# encode them
image_processing = DeformableDetrImageProcessor(format="coco_panoptic")
image_processing = image_processing_class(format="coco_panoptic")
encoding = image_processing(
images=images,
annotations=annotations,
@@ -500,10 +502,11 @@ class DeformableDetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessi
# Copied from tests.models.detr.test_image_processing_detr.DetrImageProcessingTest.test_max_width_max_height_resizing_and_pad_strategy with Detr->DeformableDetr
def test_max_width_max_height_resizing_and_pad_strategy(self):
for image_processing_class in self.image_processor_list:
image_1 = torch.ones([200, 100, 3], dtype=torch.uint8)
# do_pad=False, max_height=100, max_width=100, image=200x100 -> 100x50
image_processor = DeformableDetrImageProcessor(
image_processor = image_processing_class(
size={"max_height": 100, "max_width": 100},
do_pad=False,
)
@@ -511,21 +514,21 @@ class DeformableDetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessi
self.assertEqual(inputs["pixel_values"].shape, torch.Size([1, 3, 100, 50]))
# do_pad=False, max_height=300, max_width=100, image=200x100 -> 200x100
image_processor = DeformableDetrImageProcessor(
image_processor = image_processing_class(
size={"max_height": 300, "max_width": 100},
do_pad=False,
)
inputs = image_processor(images=[image_1], return_tensors="pt")
# do_pad=True, max_height=100, max_width=100, image=200x100 -> 100x100
image_processor = DeformableDetrImageProcessor(
image_processor = image_processing_class(
size={"max_height": 100, "max_width": 100}, do_pad=True, pad_size={"height": 100, "width": 100}
)
inputs = image_processor(images=[image_1], return_tensors="pt")
self.assertEqual(inputs["pixel_values"].shape, torch.Size([1, 3, 100, 100]))
# do_pad=True, max_height=300, max_width=100, image=200x100 -> 300x100
image_processor = DeformableDetrImageProcessor(
image_processor = image_processing_class(
size={"max_height": 300, "max_width": 100},
do_pad=True,
pad_size={"height": 301, "width": 101},
@@ -537,7 +540,7 @@ class DeformableDetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessi
image_2 = torch.ones([100, 150, 3], dtype=torch.uint8)
# do_pad=True, max_height=150, max_width=100, images=[200x100, 100x150] -> 150x100
image_processor = DeformableDetrImageProcessor(
image_processor = image_processing_class(
size={"max_height": 150, "max_width": 100},
do_pad=True,
pad_size={"height": 150, "width": 100},

181
tests/models/detr/test_image_processing_detr.py
View File

@@ -19,8 +19,8 @@ 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
from transformers.testing_utils import require_torch, require_torch_gpu, require_vision, slow
from transformers.utils import is_torch_available, is_torchvision_available, is_vision_available
from ...test_image_processing_common import AnnotationFormatTestMixin, ImageProcessingTestMixin, prepare_image_inputs
@@ -33,6 +33,9 @@ if is_vision_available():
from transformers import DetrImageProcessor
if is_torchvision_available():
from transformers import DetrImageProcessorFast
class DetrImageProcessingTester(unittest.TestCase):
def __init__(
@@ -51,6 +54,7 @@ class DetrImageProcessingTester(unittest.TestCase):
image_std=[0.5, 0.5, 0.5],
do_pad=True,
):
super().__init__()
# by setting size["longest_edge"] > max_resolution we're effectively not testing this :p
size = size if size is not None else {"shortest_edge": 18, "longest_edge": 1333}
self.parent = parent
@@ -132,6 +136,7 @@ class DetrImageProcessingTester(unittest.TestCase):
@require_vision
class DetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixin, unittest.TestCase):
image_processing_class = DetrImageProcessor if is_vision_available() else None
fast_image_processing_class = DetrImageProcessorFast if is_torchvision_available() else None
def setUp(self):
super().setUp()
@@ -142,7 +147,8 @@ class DetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixi
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)
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"))
@@ -153,11 +159,12 @@ class DetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixi
self.assertTrue(hasattr(image_processing, "do_pad"))
def test_image_processor_from_dict_with_kwargs(self):
image_processor = self.image_processing_class.from_dict(self.image_processor_dict)
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, {"shortest_edge": 18, "longest_edge": 1333})
self.assertEqual(image_processor.do_pad, True)
image_processor = self.image_processing_class.from_dict(
image_processor = image_processing_class.from_dict(
self.image_processor_dict, size=42, max_size=84, pad_and_return_pixel_mask=False
)
self.assertEqual(image_processor.size, {"shortest_edge": 42, "longest_edge": 84})
@@ -178,7 +185,8 @@ class DetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixi
}
image_processor_params = {**image_processor_dict, **{"format": "_INVALID_FORMAT_"}}
image_processor = self.image_processing_class(**image_processor_params)
for image_processing_class in self.image_processor_list:
image_processor = image_processing_class(**image_processor_params)
with self.assertRaises(ValueError) as e:
image_processor(**params)
@@ -193,8 +201,9 @@ class DetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixi
params = {"image_id": 39769, "annotations": target}
for image_processing_class in self.image_processor_list:
# encode them
image_processing = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
image_processing = image_processing_class.from_pretrained("facebook/detr-resnet-50")
# legal encodings (single image)
_ = image_processing(images=image, annotations=params, return_tensors="pt")
@@ -229,8 +238,9 @@ class DetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixi
target = {"image_id": 39769, "annotations": target}
for image_processing_class in self.image_processor_list:
# encode them
image_processing = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
image_processing = image_processing_class.from_pretrained("facebook/detr-resnet-50")
encoding = image_processing(images=image, annotations=target, return_tensors="pt")
# verify pixel values
@@ -275,8 +285,9 @@ class DetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixi
masks_path = pathlib.Path("./tests/fixtures/tests_samples/COCO/coco_panoptic")
for image_processing_class in self.image_processor_list:
# encode them
image_processing = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50-panoptic")
image_processing = image_processing_class.from_pretrained("facebook/detr-resnet-50-panoptic")
encoding = image_processing(images=image, annotations=target, masks_path=masks_path, return_tensors="pt")
# verify pixel values
@@ -305,7 +316,8 @@ class DetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixi
self.assertTrue(torch.allclose(encoding["labels"][0]["class_labels"], expected_class_labels))
# verify masks
expected_masks_sum = 822873
self.assertEqual(encoding["labels"][0]["masks"].sum().item(), expected_masks_sum)
relative_error = torch.abs(encoding["labels"][0]["masks"].sum() - expected_masks_sum) / expected_masks_sum
self.assertTrue(relative_error < 1e-3)
# verify orig_size
expected_orig_size = torch.tensor([480, 640])
self.assertTrue(torch.allclose(encoding["labels"][0]["orig_size"], expected_orig_size))
@@ -340,7 +352,8 @@ class DetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixi
images = [image_0, image_1]
annotations = [annotations_0, annotations_1]
image_processing = DetrImageProcessor()
for image_processing_class in self.image_processor_list:
image_processing = image_processing_class()
encoding = image_processing(
images=images,
annotations=annotations,
@@ -459,8 +472,9 @@ class DetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixi
images = [image_0, image_1]
annotations = [annotation_0, annotation_1]
for image_processing_class in self.image_processor_list:
# encode them
image_processing = DetrImageProcessor(format="coco_panoptic")
image_processing = image_processing_class(format="coco_panoptic")
encoding = image_processing(
images=images,
annotations=annotations,
@@ -554,10 +568,11 @@ class DetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixi
self.assertTrue(torch.allclose(encoding["labels"][1]["boxes"], expected_boxes_1, rtol=1))
def test_max_width_max_height_resizing_and_pad_strategy(self):
for image_processing_class in self.image_processor_list:
image_1 = torch.ones([200, 100, 3], dtype=torch.uint8)
# do_pad=False, max_height=100, max_width=100, image=200x100 -> 100x50
image_processor = DetrImageProcessor(
image_processor = image_processing_class(
size={"max_height": 100, "max_width": 100},
do_pad=False,
)
@@ -565,21 +580,21 @@ class DetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixi
self.assertEqual(inputs["pixel_values"].shape, torch.Size([1, 3, 100, 50]))
# do_pad=False, max_height=300, max_width=100, image=200x100 -> 200x100
image_processor = DetrImageProcessor(
image_processor = image_processing_class(
size={"max_height": 300, "max_width": 100},
do_pad=False,
)
inputs = image_processor(images=[image_1], return_tensors="pt")
# do_pad=True, max_height=100, max_width=100, image=200x100 -> 100x100
image_processor = DetrImageProcessor(
image_processor = image_processing_class(
size={"max_height": 100, "max_width": 100}, do_pad=True, pad_size={"height": 100, "width": 100}
)
inputs = image_processor(images=[image_1], return_tensors="pt")
self.assertEqual(inputs["pixel_values"].shape, torch.Size([1, 3, 100, 100]))
# do_pad=True, max_height=300, max_width=100, image=200x100 -> 300x100
image_processor = DetrImageProcessor(
image_processor = image_processing_class(
size={"max_height": 300, "max_width": 100},
do_pad=True,
pad_size={"height": 301, "width": 101},
@@ -591,7 +606,7 @@ class DetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixi
image_2 = torch.ones([100, 150, 3], dtype=torch.uint8)
# do_pad=True, max_height=150, max_width=100, images=[200x100, 100x150] -> 150x100
image_processor = DetrImageProcessor(
image_processor = image_processing_class(
size={"max_height": 150, "max_width": 100},
do_pad=True,
pad_size={"height": 150, "width": 100},
@@ -600,11 +615,12 @@ class DetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixi
self.assertEqual(inputs["pixel_values"].shape, torch.Size([2, 3, 150, 100]))
def test_longest_edge_shortest_edge_resizing_strategy(self):
for image_processing_class in self.image_processor_list:
image_1 = torch.ones([958, 653, 3], dtype=torch.uint8)
# max size is set; width < height;
# do_pad=False, longest_edge=640, shortest_edge=640, image=958x653 -> 640x436
image_processor = DetrImageProcessor(
image_processor = image_processing_class(
size={"longest_edge": 640, "shortest_edge": 640},
do_pad=False,
)
@@ -614,7 +630,7 @@ class DetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixi
image_2 = torch.ones([653, 958, 3], dtype=torch.uint8)
# max size is set; height < width;
# do_pad=False, longest_edge=640, shortest_edge=640, image=653x958 -> 436x640
image_processor = DetrImageProcessor(
image_processor = image_processing_class(
size={"longest_edge": 640, "shortest_edge": 640},
do_pad=False,
)
@@ -624,7 +640,7 @@ class DetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixi
image_3 = torch.ones([100, 120, 3], dtype=torch.uint8)
# max size is set; width == size; height > max_size;
# do_pad=False, longest_edge=118, shortest_edge=100, image=120x100 -> 118x98
image_processor = DetrImageProcessor(
image_processor = image_processing_class(
size={"longest_edge": 118, "shortest_edge": 100},
do_pad=False,
)
@@ -634,7 +650,7 @@ class DetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixi
image_4 = torch.ones([128, 50, 3], dtype=torch.uint8)
# max size is set; height == size; width < max_size;
# do_pad=False, longest_edge=256, shortest_edge=50, image=50x128 -> 50x128
image_processor = DetrImageProcessor(
image_processor = image_processing_class(
size={"longest_edge": 256, "shortest_edge": 50},
do_pad=False,
)
@@ -644,9 +660,130 @@ class DetrImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixi
image_5 = torch.ones([50, 50, 3], dtype=torch.uint8)
# max size is set; height == width; width < max_size;
# do_pad=False, longest_edge=117, shortest_edge=50, image=50x50 -> 50x50
image_processor = DetrImageProcessor(
image_processor = image_processing_class(
size={"longest_edge": 117, "shortest_edge": 50},
do_pad=False,
)
inputs = image_processor(images=[image_5], return_tensors="pt")
self.assertEqual(inputs["pixel_values"].shape, torch.Size([1, 3, 50, 50]))
@slow
@require_torch_gpu
def test_fast_processor_equivalence_cpu_gpu_coco_detection_annotations(self):
# prepare image and target
image = Image.open("./tests/fixtures/tests_samples/COCO/000000039769.png")
with open("./tests/fixtures/tests_samples/COCO/coco_annotations.txt", "r") as f:
target = json.loads(f.read())
target = {"image_id": 39769, "annotations": target}
processor = self.image_processor_list[1].from_pretrained("facebook/detr-resnet-50")
# 1. run processor on CPU
encoding_cpu = processor(images=image, annotations=target, return_tensors="pt", device="cpu")
# 2. run processor on GPU
encoding_gpu = processor(images=image, annotations=target, return_tensors="pt", device="cuda")
# verify pixel values
self.assertEqual(encoding_cpu["pixel_values"].shape, encoding_gpu["pixel_values"].shape)
self.assertTrue(
torch.allclose(
encoding_cpu["pixel_values"][0, 0, 0, :3],
encoding_gpu["pixel_values"][0, 0, 0, :3].to("cpu"),
atol=1e-4,
)
)
# verify area
self.assertTrue(torch.allclose(encoding_cpu["labels"][0]["area"], encoding_gpu["labels"][0]["area"].to("cpu")))
# verify boxes
self.assertEqual(encoding_cpu["labels"][0]["boxes"].shape, encoding_gpu["labels"][0]["boxes"].shape)
self.assertTrue(
torch.allclose(
encoding_cpu["labels"][0]["boxes"][0], encoding_gpu["labels"][0]["boxes"][0].to("cpu"), atol=1e-3
)
)
# verify image_id
self.assertTrue(
torch.allclose(encoding_cpu["labels"][0]["image_id"], encoding_gpu["labels"][0]["image_id"].to("cpu"))
)
# verify is_crowd
self.assertTrue(
torch.allclose(encoding_cpu["labels"][0]["iscrowd"], encoding_gpu["labels"][0]["iscrowd"].to("cpu"))
)
# verify class_labels
self.assertTrue(
torch.allclose(
encoding_cpu["labels"][0]["class_labels"], encoding_gpu["labels"][0]["class_labels"].to("cpu")
)
)
# verify orig_size
self.assertTrue(
torch.allclose(encoding_cpu["labels"][0]["orig_size"], encoding_gpu["labels"][0]["orig_size"].to("cpu"))
)
# verify size
self.assertTrue(torch.allclose(encoding_cpu["labels"][0]["size"], encoding_gpu["labels"][0]["size"].to("cpu")))
@slow
@require_torch_gpu
def test_fast_processor_equivalence_cpu_gpu_coco_panoptic_annotations(self):
# prepare image, target and masks_path
image = Image.open("./tests/fixtures/tests_samples/COCO/000000039769.png")
with open("./tests/fixtures/tests_samples/COCO/coco_panoptic_annotations.txt", "r") as f:
target = json.loads(f.read())
target = {"file_name": "000000039769.png", "image_id": 39769, "segments_info": target}
masks_path = pathlib.Path("./tests/fixtures/tests_samples/COCO/coco_panoptic")
processor = self.image_processor_list[1].from_pretrained("facebook/detr-resnet-50-panoptic")
# 1. run processor on CPU
encoding_cpu = processor(
images=image, annotations=target, masks_path=masks_path, return_tensors="pt", device="cpu"
)
# 2. run processor on GPU
encoding_gpu = processor(
images=image, annotations=target, masks_path=masks_path, return_tensors="pt", device="cuda"
)
# verify pixel values
self.assertEqual(encoding_cpu["pixel_values"].shape, encoding_gpu["pixel_values"].shape)
self.assertTrue(
torch.allclose(
encoding_cpu["pixel_values"][0, 0, 0, :3],
encoding_gpu["pixel_values"][0, 0, 0, :3].to("cpu"),
atol=1e-4,
)
)
# verify area
self.assertTrue(torch.allclose(encoding_cpu["labels"][0]["area"], encoding_gpu["labels"][0]["area"].to("cpu")))
# verify boxes
self.assertEqual(encoding_cpu["labels"][0]["boxes"].shape, encoding_gpu["labels"][0]["boxes"].shape)
self.assertTrue(
torch.allclose(
encoding_cpu["labels"][0]["boxes"][0], encoding_gpu["labels"][0]["boxes"][0].to("cpu"), atol=1e-3
)
)
# verify image_id
self.assertTrue(
torch.allclose(encoding_cpu["labels"][0]["image_id"], encoding_gpu["labels"][0]["image_id"].to("cpu"))
)
# verify is_crowd
self.assertTrue(
torch.allclose(encoding_cpu["labels"][0]["iscrowd"], encoding_gpu["labels"][0]["iscrowd"].to("cpu"))
)
# verify class_labels
self.assertTrue(
torch.allclose(
encoding_cpu["labels"][0]["class_labels"], encoding_gpu["labels"][0]["class_labels"].to("cpu")
)
)
# verify masks
masks_sum_cpu = encoding_cpu["labels"][0]["masks"].sum()
masks_sum_gpu = encoding_gpu["labels"][0]["masks"].sum()
relative_error = torch.abs(masks_sum_cpu - masks_sum_gpu) / masks_sum_cpu
self.assertTrue(relative_error < 1e-3)
# verify orig_size
self.assertTrue(
torch.allclose(encoding_cpu["labels"][0]["orig_size"], encoding_gpu["labels"][0]["orig_size"].to("cpu"))
)
# verify size
self.assertTrue(torch.allclose(encoding_cpu["labels"][0]["size"], encoding_gpu["labels"][0]["size"].to("cpu")))

17
tests/models/grounding_dino/test_image_processing_grounding_dino.py
View File

@@ -269,7 +269,8 @@ class GroundingDinoImageProcessingTest(AnnotationFormatTestMixin, ImageProcessin
images = [image_0, image_1]
annotations = [annotations_0, annotations_1]
image_processing = GroundingDinoImageProcessor()
for image_processing_class in self.image_processor_list:
image_processing = image_processing_class()
encoding = image_processing(
images=images,
annotations=annotations,
@@ -440,8 +441,9 @@ class GroundingDinoImageProcessingTest(AnnotationFormatTestMixin, ImageProcessin
images = [image_0, image_1]
annotations = [annotation_0, annotation_1]
for image_processing_class in self.image_processor_list:
# encode them
image_processing = GroundingDinoImageProcessor(format="coco_panoptic")
image_processing = image_processing_class(format="coco_panoptic")
encoding = image_processing(
images=images,
annotations=annotations,
@@ -536,10 +538,11 @@ class GroundingDinoImageProcessingTest(AnnotationFormatTestMixin, ImageProcessin
# Copied from tests.models.detr.test_image_processing_detr.DetrImageProcessingTest.test_max_width_max_height_resizing_and_pad_strategy with Detr->GroundingDino
def test_max_width_max_height_resizing_and_pad_strategy(self):
for image_processing_class in self.image_processor_list:
image_1 = torch.ones([200, 100, 3], dtype=torch.uint8)
# do_pad=False, max_height=100, max_width=100, image=200x100 -> 100x50
image_processor = GroundingDinoImageProcessor(
image_processor = image_processing_class(
size={"max_height": 100, "max_width": 100},
do_pad=False,
)
@@ -547,21 +550,21 @@ class GroundingDinoImageProcessingTest(AnnotationFormatTestMixin, ImageProcessin
self.assertEqual(inputs["pixel_values"].shape, torch.Size([1, 3, 100, 50]))
# do_pad=False, max_height=300, max_width=100, image=200x100 -> 200x100
image_processor = GroundingDinoImageProcessor(
image_processor = image_processing_class(
size={"max_height": 300, "max_width": 100},
do_pad=False,
)
inputs = image_processor(images=[image_1], return_tensors="pt")
# do_pad=True, max_height=100, max_width=100, image=200x100 -> 100x100
image_processor = GroundingDinoImageProcessor(
image_processor = image_processing_class(
size={"max_height": 100, "max_width": 100}, do_pad=True, pad_size={"height": 100, "width": 100}
)
inputs = image_processor(images=[image_1], return_tensors="pt")
self.assertEqual(inputs["pixel_values"].shape, torch.Size([1, 3, 100, 100]))
# do_pad=True, max_height=300, max_width=100, image=200x100 -> 300x100
image_processor = GroundingDinoImageProcessor(
image_processor = image_processing_class(
size={"max_height": 300, "max_width": 100},
do_pad=True,
pad_size={"height": 301, "width": 101},
@@ -573,7 +576,7 @@ class GroundingDinoImageProcessingTest(AnnotationFormatTestMixin, ImageProcessin
image_2 = torch.ones([100, 150, 3], dtype=torch.uint8)
# do_pad=True, max_height=150, max_width=100, images=[200x100, 100x150] -> 150x100
image_processor = GroundingDinoImageProcessor(
image_processor = image_processing_class(
size={"max_height": 150, "max_width": 100},
do_pad=True,
pad_size={"height": 150, "width": 100},

11
tests/models/yolos/test_image_processing_yolos.py
View File

@@ -553,10 +553,11 @@ class YolosImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMix
# Copied from tests.models.detr.test_image_processing_detr.DetrImageProcessingTest.test_max_width_max_height_resizing_and_pad_strategy with Detr->Yolos
def test_max_width_max_height_resizing_and_pad_strategy(self):
for image_processing_class in self.image_processor_list:
image_1 = torch.ones([200, 100, 3], dtype=torch.uint8)
# do_pad=False, max_height=100, max_width=100, image=200x100 -> 100x50
image_processor = YolosImageProcessor(
image_processor = image_processing_class(
size={"max_height": 100, "max_width": 100},
do_pad=False,
)
@@ -564,21 +565,21 @@ class YolosImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMix
self.assertEqual(inputs["pixel_values"].shape, torch.Size([1, 3, 100, 50]))
# do_pad=False, max_height=300, max_width=100, image=200x100 -> 200x100
image_processor = YolosImageProcessor(
image_processor = image_processing_class(
size={"max_height": 300, "max_width": 100},
do_pad=False,
)
inputs = image_processor(images=[image_1], return_tensors="pt")
# do_pad=True, max_height=100, max_width=100, image=200x100 -> 100x100
image_processor = YolosImageProcessor(
image_processor = image_processing_class(
size={"max_height": 100, "max_width": 100}, do_pad=True, pad_size={"height": 100, "width": 100}
)
inputs = image_processor(images=[image_1], return_tensors="pt")
self.assertEqual(inputs["pixel_values"].shape, torch.Size([1, 3, 100, 100]))
# do_pad=True, max_height=300, max_width=100, image=200x100 -> 300x100
image_processor = YolosImageProcessor(
image_processor = image_processing_class(
size={"max_height": 300, "max_width": 100},
do_pad=True,
pad_size={"height": 301, "width": 101},
@@ -590,7 +591,7 @@ class YolosImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMix
image_2 = torch.ones([100, 150, 3], dtype=torch.uint8)
# do_pad=True, max_height=150, max_width=100, images=[200x100, 100x150] -> 150x100
image_processor = YolosImageProcessor(
image_processor = image_processing_class(
size={"max_height": 150, "max_width": 100},
do_pad=True,
pad_size={"height": 150, "width": 100},

2
tests/test_image_processing_common.py
View File

@@ -191,7 +191,7 @@ class ImageProcessingTestMixin:
dummy_images = torch.randint(0, 255, (4, 3, 224, 224), dtype=torch.uint8)
image_processor_slow = self.image_processing_class(**self.image_processor_dict)
image_processor_fast = self.fast_image_processing_class()
image_processor_fast = self.fast_image_processing_class(**self.image_processor_dict)
fast_time = measure_time(image_processor_fast, dummy_images)
slow_time = measure_time(image_processor_slow, dummy_images)