def test_build_dict_field_transform_default_imagenet(self):
dataset = self.get_test_image_dataset(SampleType.DICT)
# should apply the transform in the config
config = [{"name": "ToTensor"}]
default_transform = transforms.Compose(
[transforms.CenterCrop(100), transforms.ToTensor()]
)
transform = build_field_transform_default_imagenet(
config, default_transform=default_transform
)
sample = dataset[0]
expected_sample = _apply_transform_to_key_and_copy(
sample, transforms.ToTensor(), "input"
)
self.transform_checks(sample, transform, expected_sample)
# should apply default_transform
config = None
transform = build_field_transform_default_imagenet(
config, default_transform=default_transform
)
expected_sample = _apply_transform_to_key_and_copy(
sample, default_transform, "input"
)
self.transform_checks(sample, transform, expected_sample)
# should apply the transform for a test split
transform = build_field_transform_default_imagenet(config, split="test")
expected_sample = _apply_transform_to_key_and_copy(
sample, ImagenetNoAugmentTransform(), "input"
)
self.transform_checks(sample, transform, expected_sample)
def test_transforms(self):
input = self.get_test_image()
# reference transform which we will use to validate the built transforms
reference_transform = ImagenetNoAugmentTransform()
reference_output = reference_transform(input)
# test a registered transform
config = [{"name": "imagenet_no_augment"}]
transform = build_transforms(config)
output = transform(input)
self.assertTrue(torch.allclose(output, reference_output))
# test a transform built using torchvision transforms
config = [
{"name": "Resize", "size": 256},
{"name": "CenterCrop", "size": 224},
{"name": "ToTensor"},
{
"name": "Normalize",
"mean": [0.485, 0.456, 0.406],
"std": [0.229, 0.224, 0.225],
},
]
transform = build_transforms(config)
output = transform(input)
self.assertTrue(torch.allclose(output, reference_output))
# test a combination of registered and torchvision transforms
config = [
{"name": "resize", "size": 256},
{"name": "center_crop", "size": 224},
{"name": "ToTensor"},
{
"name": "Normalize",
"mean": [0.485, 0.456, 0.406],
"std": [0.229, 0.224, 0.225],
},
]
transform = build_transforms(config)
output = transform(input)
self.assertTrue(torch.allclose(output, reference_output))
def test_generic_image_transform(self):
dataset = self.get_test_image_dataset(SampleType.TUPLE)
# Check constructor asserts
with self.assertRaises(AssertionError):
transform = GenericImageTransform(
split="train", transform=transforms.ToTensor()
)
transform = GenericImageTransform(split="valid", transform=None)
# Check class constructor
transform = GenericImageTransform(transform=None)
PIL_sample = dataset[0]
tensor_sample = (transforms.ToTensor()(PIL_sample[0]), PIL_sample[1])
expected_sample = {
"input": copy.deepcopy(tensor_sample[0]),
"target": copy.deepcopy(tensor_sample[1]),
}
self.transform_checks(tensor_sample, transform, expected_sample)
transform = GenericImageTransform(transform=transforms.ToTensor())
sample = dataset[0]
expected_sample = _apply_transform_to_key_and_copy(
{"input": sample[0], "target": sample[1]}, transforms.ToTensor(), "input"
)
self.transform_checks(sample, transform, expected_sample)
transform = GenericImageTransform(split="train")
sample = dataset[0]
expected_sample = _apply_transform_to_key_and_copy(
{"input": sample[0], "target": sample[1]},
ImagenetAugmentTransform(),
"input",
)
self.transform_checks(sample, transform, expected_sample)
transform = GenericImageTransform(split="test")
sample = dataset[0]
expected_sample = _apply_transform_to_key_and_copy(
{"input": sample[0], "target": sample[1]},
ImagenetNoAugmentTransform(),
"input",
)
self.transform_checks(sample, transform, expected_sample)
# Check from_config constructor / registry
config = [
{"name": "generic_image_transform", "transforms": [{"name": "ToTensor"}]}
]
transform = build_transforms(config)
sample = dataset[0]
expected_sample = _apply_transform_to_key_and_copy(
{"input": sample[0], "target": sample[1]}, transforms.ToTensor(), "input"
)
self.transform_checks(sample, transform, expected_sample)
# Check with Imagenet defaults
config = [{"name": "generic_image_transform", "split": "train"}]
transform = build_transforms(config)
sample = dataset[0]
expected_sample = _apply_transform_to_key_and_copy(
{"input": sample[0], "target": sample[1]},
ImagenetAugmentTransform(),
"input",
)
self.transform_checks(sample, transform, expected_sample)
config = [{"name": "generic_image_transform", "split": "test"}]
transform = build_transforms(config)
sample = dataset[0]
expected_sample = _apply_transform_to_key_and_copy(
{"input": sample[0], "target": sample[1]},
ImagenetNoAugmentTransform(),
"input",
)
self.transform_checks(sample, transform, expected_sample)