more tests about the transforms with MetaTensor (#4521)

* at least scale one

Signed-off-by: Wenqi Li <[email protected]>

* add more tests

Signed-off-by: Wenqi Li <[email protected]>

* dict-based tests

Signed-off-by: Wenqi Li <[email protected]>

* reviewed intensity transforms

Signed-off-by: Wenqi Li <[email protected]>

* review post

Signed-off-by: Wenqi Li <[email protected]>

* review utils

Signed-off-by: Wenqi Li <[email protected]>

* integration tests

Signed-off-by: Wenqi Li <[email protected]>

* adds pixdim property, fixes typos

Signed-off-by: Wenqi Li <[email protected]>

Author: wyli

monai/data/__init__.py

@@ -71,6 +71,7 @@
 from .thread_buffer import ThreadBuffer, ThreadDataLoader
 from .torchscript_utils import load_net_with_metadata, save_net_with_metadata
 from .utils import (
+    affine_to_spacing,
     compute_importance_map,
     compute_shape_offset,
     convert_tables_to_dicts,

monai/data/meta_tensor.py

@@ -19,7 +19,7 @@
 
 from monai.config.type_definitions import NdarrayTensor
 from monai.data.meta_obj import MetaObj, get_track_meta
-from monai.data.utils import decollate_batch, list_data_collate, remove_extra_metadata
+from monai.data.utils import affine_to_spacing, decollate_batch, list_data_collate, remove_extra_metadata
 from monai.utils.enums import PostFix
 from monai.utils.type_conversion import convert_to_tensor
 
@@ -304,13 +304,18 @@ def as_dict(self, key: str) -> dict:
     @property
     def affine(self) -> torch.Tensor:
         """Get the affine."""
-        return self.meta["affine"]  # type: ignore
+        return self.meta.get("affine", self.get_default_affine())  # type: ignore
 
     @affine.setter
     def affine(self, d: NdarrayTensor) -> None:
         """Set the affine."""
         self.meta["affine"] = torch.as_tensor(d, device=self.device)
 
+    @property
+    def pixdim(self):
+        """Get the spacing"""
+        return affine_to_spacing(self.affine)
+
     def new_empty(self, size, dtype=None, device=None, requires_grad=False):
         """
         must be defined for deepcopy to work

monai/transforms/intensity/array.py

@@ -71,6 +71,7 @@
     "HistogramNormalize",
     "IntensityRemap",
     "RandIntensityRemap",
+    "ForegroundMask",
 ]
 
 

monai/transforms/post/dictionary.py

@@ -645,17 +645,17 @@ def __call__(self, data: Mapping[Hashable, Any]) -> Dict[Hashable, Any]:
                     trans_info=deepcopy(transform_info), mode="nearest", align_corners=None
                 )
 
-            input = d[key]
-            if isinstance(input, torch.Tensor):
-                input = input.detach()
+            inputs = d[key]
+            if isinstance(inputs, torch.Tensor):
+                inputs = inputs.detach()
 
-            if not isinstance(input, MetaTensor):
-                input = MetaTensor(input)
-            input.applied_operations = transform_info
-            input.meta = meta_info
+            if not isinstance(inputs, MetaTensor):
+                inputs = MetaTensor(inputs)
+            inputs.applied_operations = transform_info
+            inputs.meta = meta_info
 
             # construct the input dict data
-            input_dict = {orig_key: input}
+            input_dict = {orig_key: inputs}
 
             with allow_missing_keys_mode(self.transform):  # type: ignore
                 inverted = self.transform.inverse(input_dict)

monai/transforms/spatial/array.py

@@ -345,10 +345,10 @@ class ResampleToMatch(SpatialResample):
     and the size of the output image will match."""
 
     @deprecated_arg(
-        name="src_meta", since="0.8", msg_suffix="img should be `MetaTensor`, so affine can be extacted directly."
+        name="src_meta", since="0.8", msg_suffix="img should be `MetaTensor`, so affine can be extracted directly."
     )
     @deprecated_arg(
-        name="dst_meta", since="0.8", msg_suffix="img_dst should be `MetaTensor`, so affine can be extacted directly."
+        name="dst_meta", since="0.8", msg_suffix="img_dst should be `MetaTensor`, so affine can be extracted directly."
     )
     def __call__(  # type: ignore
         self,

monai/transforms/utility/dictionary.py

@@ -1011,6 +1011,7 @@ class Lambdad(MapTransform, InvertibleTransform):
         print(lambd(input_data)['label'].shape)
         (4, 2, 2)
 
+
     Args:
         keys: keys of the corresponding items to be transformed.
             See also: :py:class:`monai.transforms.compose.MapTransform`

monai/transforms/utils.py

@@ -1589,7 +1589,7 @@ def scale_affine(affine, spatial_size, new_spatial_size, centered: bool = True):
     if spatial_size == new_spatial_size:
         return affine
     r = len(affine) - 1
-    s = np.array([float(o) / max(n, 0) for o, n in zip(spatial_size, new_spatial_size)])
+    s = np.array([float(o) / float(max(n, 1)) for o, n in zip(spatial_size, new_spatial_size)])
     scale = create_scale(r, s.tolist())
     if centered:
         scale[:r, -1] = (np.diag(scale)[:r] - 1) / 2  # type: ignore

tests/test_activations.py

@@ -76,7 +76,7 @@
 
 
 class TestActivations(unittest.TestCase):
-    @parameterized.expand(TEST_CASES[:3])
+    @parameterized.expand(TEST_CASES)
     def test_value_shape(self, input_param, img, out, expected_shape):
         result = Activations(**input_param)(img)
 

tests/test_adjust_contrast.py

@@ -29,7 +29,9 @@ class TestAdjustContrast(NumpyImageTestCase2D):
     def test_correct_results(self, gamma):
         adjuster = AdjustContrast(gamma=gamma)
         for p in TEST_NDARRAYS:
-            result = adjuster(p(self.imt))
+            im = p(self.imt)
+            result = adjuster(im)
+            self.assertTrue(type(im), type(result))
             if gamma == 1.0:
                 expected = self.imt
             else:

tests/test_as_channel_last.py

@@ -29,6 +29,7 @@ class TestAsChannelLast(unittest.TestCase):
     def test_shape(self, in_type, input_param, expected_shape):
         test_data = in_type(np.random.randint(0, 2, size=[1, 2, 3, 4]))
         result = AsChannelLast(**input_param)(test_data)
+        self.assertEqual(type(result), type(test_data))
         self.assertTupleEqual(result.shape, expected_shape)