def __init__(self, keys: KeysCollection) -> None:
"""
Args:
keys: keys of the corresponding items to be transformed.
See also: :py:class:`monai.transforms.compose.MapTransform`
"""
super().__init__(keys)
self.converter = ToNumpy()
def __init__(self, keys: KeysCollection, allow_missing_keys: bool = False) -> None:
"""
Args:
keys: keys of the corresponding items to be transformed.
See also: :py:class:`monai.transforms.compose.MapTransform`
allow_missing_keys: don't raise exception if key is missing.
"""
super().__init__(keys, allow_missing_keys)
self.converter = ToNumpy()
def inverse(self, data: Mapping[Hashable, Any]) -> Dict[Hashable, Any]:
d = deepcopy(dict(data))
for key in self.key_iterator(d):
# Create inverse transform
inverse_transform = ToNumpy()
# Apply inverse
d[key] = inverse_transform(d[key])
# Remove the applied transform
self.pop_transform(d, key)
return d
def test_dataset(self):
with tempfile.TemporaryDirectory() as tempdir:
full_names, ref_data = [], []
for filename in FILENAMES:
test_image = np.random.randint(0, 2, size=(4, 4, 4))
ref_data.append(test_image)
save_path = os.path.join(tempdir, filename)
full_names.append(save_path)
nib.save(nib.Nifti1Image(test_image, np.eye(4)), save_path)
# default loading no meta
dataset = ImageDataset(full_names)
for d, ref in zip(dataset, ref_data):
np.testing.assert_allclose(d, ref, atol=1e-3)
# loading no meta, int
dataset = ImageDataset(full_names, dtype=np.float16)
for d, _ in zip(dataset, ref_data):
self.assertEqual(d.dtype, np.float16)
# loading with meta, no transform
dataset = ImageDataset(full_names, image_only=False)
for d_tuple, ref in zip(dataset, ref_data):
d, meta = d_tuple
np.testing.assert_allclose(d, ref, atol=1e-3)
np.testing.assert_allclose(meta["original_affine"], np.eye(4))
# loading image/label, no meta
dataset = ImageDataset(full_names, seg_files=full_names, image_only=True)
for d_tuple, ref in zip(dataset, ref_data):
img, seg = d_tuple
np.testing.assert_allclose(img, ref, atol=1e-3)
np.testing.assert_allclose(seg, ref, atol=1e-3)
# loading image/label, no meta
dataset = ImageDataset(full_names, transform=lambda x: x + 1, image_only=True)
for d, ref in zip(dataset, ref_data):
np.testing.assert_allclose(d, ref + 1, atol=1e-3)
# loading image/label, with meta
dataset = ImageDataset(
full_names,
transform=lambda x: x + 1,
seg_files=full_names,
seg_transform=lambda x: x + 2,
image_only=False,
)
for d_tuple, ref in zip(dataset, ref_data):
img, seg, meta, seg_meta = d_tuple
np.testing.assert_allclose(img, ref + 1, atol=1e-3)
np.testing.assert_allclose(seg, ref + 2, atol=1e-3)
np.testing.assert_allclose(meta["original_affine"], np.eye(4), atol=1e-3)
np.testing.assert_allclose(seg_meta["original_affine"], np.eye(4), atol=1e-3)
# loading image/label, with meta
dataset = ImageDataset(
image_files=full_names,
seg_files=full_names,
labels=[1, 2, 3],
transform=lambda x: x + 1,
label_transform=Compose(
[
ToNumpy(),
MapLabelValue(orig_labels=[1, 2, 3], target_labels=[30.0, 20.0, 10.0], dtype=np.float32),
]
),
image_only=False,
)
for idx, (d_tuple, ref) in enumerate(zip(dataset, ref_data)):
img, seg, label, meta, seg_meta = d_tuple
np.testing.assert_allclose(img, ref + 1, atol=1e-3)
np.testing.assert_allclose(seg, ref, atol=1e-3)
# test label_transform
np.testing.assert_allclose((3 - idx) * 10.0, label)
self.assertTrue(isinstance(label, np.ndarray))
self.assertEqual(label.dtype, np.float32)
np.testing.assert_allclose(meta["original_affine"], np.eye(4), atol=1e-3)
np.testing.assert_allclose(seg_meta["original_affine"], np.eye(4), atol=1e-3)
# loading image/label, with sync. transform
dataset = ImageDataset(
full_names, transform=RandTest(), seg_files=full_names, seg_transform=RandTest(), image_only=False
)
for d_tuple, ref in zip(dataset, ref_data):
img, seg, meta, seg_meta = d_tuple
np.testing.assert_allclose(img, seg, atol=1e-3)
self.assertTrue(not np.allclose(img, ref))
np.testing.assert_allclose(meta["original_affine"], np.eye(4), atol=1e-3)