Always convert input to C-order in distance_transform_edt

monai/transforms/utils.py

@@ -2190,7 +2190,7 @@ def distance_transform_edt(
         if return_distances:
             dtype = torch.float64 if float64_distances else torch.float32
             if distances is None:
-                distances = torch.zeros_like(img, dtype=dtype)  # type: ignore
+                distances = torch.zeros_like(img, memory_format=torch.contiguous_format, dtype=dtype)  # type: ignore
             else:
                 if not isinstance(distances, torch.Tensor) and distances.device != img.device:
                     raise TypeError("distances must be a torch.Tensor on the same device as img")

tests/test_clip_intensity_percentiles.py

@@ -18,7 +18,6 @@
 from monai.transforms import ClipIntensityPercentiles
 from monai.transforms.utils import soft_clip
 from monai.transforms.utils_pytorch_numpy_unification import clip, percentile
-from monai.utils.type_conversion import convert_to_tensor
 from tests.utils import TEST_NDARRAYS, NumpyImageTestCase2D, NumpyImageTestCase3D, assert_allclose
 
 
@@ -30,7 +29,7 @@ def test_hard_clipping_two_sided(self, p):
         im = p(self.imt)
         result = hard_clipper(im)
         lower, upper = percentile(im, (5, 95))
-        expected = clip(convert_to_tensor(im), lower, upper)
+        expected = clip(im, lower, upper)
         assert_allclose(result, p(expected), type_test="tensor", rtol=1e-4, atol=0)
 
     @parameterized.expand([[p] for p in TEST_NDARRAYS])