def test_if_applies_transforms(self, input_directory):
transform = transformations.RandomCrop((2, 2))
dataset1 = datasets.NiftiFolder.from_dir(input_directory)
dataset2 = datasets.NiftiFolder.from_dir(input_directory)
dataset3 = datasets.NiftiFolder.from_dir(input_directory)
dataset = datasets.CombinedDataset(dataset1,
dataset2,
dataset3,
transform=transform)
entry = dataset[0]
for img in entry:
assert np.all(img.shape == np.array((10, 16, 2, 2)))
trfs.Lambda(lambda x: F.pad(x, [0, 0, 0, 0, 5, 0])
if x.shape[1] % 2 != 0 else x),
transformations.StandardizeVolumeWithFilter(0),
trfs.Lambda(lambda x: x.float())
])
masks_transformations = trfs.Compose([
trfs.Lambda(lambda x: np.expand_dims(x, 3)),
transformations.NiftiToTorchDimensionsReorderTransformation(),
trfs.Lambda(lambda x: torch.from_numpy(x)),
transformations.OneHotEncoding([0, 1, 2, 3]),
trfs.Lambda(lambda x: F.pad(x, [0, 0, 0, 0, 5, 0])
if x.shape[1] % 16 != 0 else x),
trfs.Lambda(lambda x: x.float())
])
common_transformations = transformations.ComposeCommon(
[transformations.RandomCrop((args.input_size, args.input_size))])
volumes_paths, masks_paths = read_dataset_json(args.dataset_json)
volumes_set = datasets.NiftiFolder(volumes_paths, volumes_transformations)
masks_set = datasets.NiftiFolder(masks_paths, masks_transformations)
combined_set = datasets.CombinedDataset(volumes_set,
masks_set,
transform=common_transformations)
with open(args.division_json, "r") as division_file:
indices = json.load(division_file)
train_set = Subset(combined_set, indices["train"])
valid_set = Subset(combined_set, indices["valid"])
train_loader = torch.utils.data.DataLoader(train_set,
batch_size=args.batch_size,
shuffle=True)
def test_if_processes_variadic_types(self, imgs):
transformation = trfs.RandomCrop((5, 5))
desired_shape = (4, 3, 5, 5)
assert all([tuple(img.shape) == desired_shape for img in transformation(*imgs)])
def test_if_processes_variadic_arguments(self, imgs):
transformation = trfs.RandomCrop((5, 5))
desired = torch.zeros((4, 3, 5, 5))
assert all([img.shape == desired.shape for img in transformation(*imgs)])
def test_if_raises_on_mismatched_shapes(self, img1, img2, img3):
transformation = trfs.RandomCrop((10, 10))
with pytest.raises(AssertionError):
transformation(img1, img2, img3)
def test_if_returns_correct_shapes(self, img, size, desired):
transformation = trfs.RandomCrop(size)
transformed = transformation(img)[0]
assert transformed.shape == desired.shape