def test_resample(self, input_param, input_data, expected_val):
warp_layer = Warp(**input_param)
result = warp_layer(**input_data)
np.testing.assert_allclose(result.cpu().numpy(),
expected_val.cpu().numpy(),
rtol=1e-4,
atol=1e-4)
def test_grad(self):
for b in GridSampleMode:
for p in GridSamplePadMode:
warp_layer = Warp(mode=b.value, padding_mode=p.value)
input_image = torch.rand(
(2, 3, 20, 20), dtype=torch.float64) * 10.0
ddf = torch.rand((2, 2, 20, 20), dtype=torch.float64) * 2.0
input_image.requires_grad = True
ddf.requires_grad = False # Jacobian mismatch for output 0 with respect to input 1
gradcheck(warp_layer, (input_image, ddf), atol=1e-2, eps=1e-2)
def test_ill_shape(self):
warp_layer = Warp(spatial_dims=2)
with self.assertRaisesRegex(ValueError, ""):
warp_layer(
image=torch.arange(4).reshape((1, 1, 1, 2, 2)).to(dtype=torch.float), ddf=torch.zeros(1, 2, 2, 2)
)
with self.assertRaisesRegex(ValueError, ""):
warp_layer(
image=torch.arange(4).reshape((1, 1, 2, 2)).to(dtype=torch.float), ddf=torch.zeros(1, 2, 1, 2, 2)
)
with self.assertRaisesRegex(ValueError, ""):
warp_layer(image=torch.arange(4).reshape((1, 1, 2, 2)).to(dtype=torch.float), ddf=torch.zeros(1, 2, 3, 3))
def monai_warp(img, ddf):
"""
warp with MONAI
Args:
img: numpy array of shape (D, H, W)
ddf: numpy array of shape (3, D, H, W)
Returns:
warped_img: numpy arrap of shape (D, H, W)
"""
warp_layer = Warp(padding_mode="zeros")
# turn to tensor and add channel dim
monai_img = torch.tensor(img).unsqueeze(0)
ddf = torch.tensor(ddf)
# img -> batch -> img
warped_img = warp_layer(monai_img.unsqueeze(0),
ddf.unsqueeze(0)).squeeze(0)
# remove channel dim
warped_img = np.asarray(warped_img.squeeze(0))
return warped_img
def test_ill_opts(self):
with self.assertRaisesRegex(ValueError, ""):
Warp(spatial_dims=4)
