def test_correct_results(self, angle, keep_size, mode, padding_mode,
align_corners):
rotate_fn = Rotated(("img", "seg"), [0, angle, 0], keep_size,
(mode, "nearest"), padding_mode, align_corners)
rotated = rotate_fn({"img": self.imt[0], "seg": self.segn[0]})
if keep_size:
np.testing.assert_allclose(self.imt[0].shape, rotated["img"].shape)
_order = 0 if mode == "nearest" else 1
if padding_mode == "border":
_mode = "nearest"
elif padding_mode == "reflection":
_mode = "reflect"
else:
_mode = "constant"
expected = scipy.ndimage.rotate(self.imt[0, 0],
np.rad2deg(angle), (0, 2),
not keep_size,
order=_order,
mode=_mode,
prefilter=False)
np.testing.assert_allclose(expected.astype(np.float32),
rotated["img"][0],
atol=1e-3)
expected = scipy.ndimage.rotate(self.segn[0, 0],
np.rad2deg(angle), (0, 2),
not keep_size,
order=0,
mode=_mode,
prefilter=False)
expected = np.stack(expected).astype(int)
self.assertLessEqual(np.count_nonzero(expected != rotated["seg"][0]),
105)
def _rotate(x, degrees: uniform = 10.0):
rotation = Rotated(keys=['image'],
angle=degrees,
spatial_axes=(0, 1),
reshape=False,
order=0,
mode='constant',
cval=0,
prefilter=True)
data_dict = {'image': x.data}
rotation_dict = rotation(data_dict)
x = rotation_dict['image'][0]
return torch.Tensor(x[None])
def test_correct_results(self, im_type, angle, keep_size, mode,
padding_mode, align_corners):
rotate_fn = Rotated(("img", "seg"), [0, angle, 0],
keep_size, (mode, "nearest"),
padding_mode,
align_corners,
dtype=np.float64)
rotated = rotate_fn({
"img": im_type(self.imt[0]),
"seg": im_type(self.segn[0])
})
if keep_size:
np.testing.assert_allclose(self.imt[0].shape, rotated["img"].shape)
_order = 0 if mode == "nearest" else 1
if padding_mode == "border":
_mode = "nearest"
elif padding_mode == "reflection":
_mode = "reflect"
else:
_mode = "constant"
expected = scipy.ndimage.rotate(self.imt[0, 0],
np.rad2deg(angle), (0, 2),
not keep_size,
order=_order,
mode=_mode,
prefilter=False)
for k, v in rotated.items():
rotated[k] = v.cpu() if isinstance(v, torch.Tensor) else v
good = np.sum(
np.isclose(expected.astype(np.float32),
rotated["img"][0],
atol=1e-3))
self.assertLessEqual(np.abs(good - expected.size), 5,
"diff at most 5 voxels.")
expected = scipy.ndimage.rotate(self.segn[0, 0],
np.rad2deg(angle), (0, 2),
not keep_size,
order=0,
mode=_mode,
prefilter=False)
expected = np.stack(expected).astype(int)
if isinstance(rotated["seg"], MetaTensor):
rotated["seg"] = rotated["seg"].as_tensor(
) # pytorch 1.7 compatible
self.assertLessEqual(np.count_nonzero(expected != rotated["seg"][0]),
160)
def test_correct_results(self, angle, spatial_axes, reshape, order, mode,
cval, prefilter):
key = "img"
rotate_fn = Rotated(key, angle, spatial_axes, reshape, order, mode,
cval, prefilter)
rotated = rotate_fn({key: self.imt[0]})
expected = list()
for channel in self.imt[0]:
expected.append(
scipy.ndimage.rotate(channel,
angle,
spatial_axes,
reshape,
order=order,
mode=mode,
cval=cval,
prefilter=prefilter))
expected = np.stack(expected).astype(np.float32)
self.assertTrue(np.allclose(expected, rotated[key]))
1, [0.5, 0.6, 0.9], [1.1, 1, 1.05],
keep_size=True)))
TESTS.append((
"RandRotated, prob 0",
"2D",
0,
RandRotated(KEYS, prob=0),
))
TESTS.append((
"Rotated 2d",
"2D",
8e-2,
Rotated(KEYS,
random.uniform(np.pi / 6, np.pi),
keep_size=True,
align_corners=False),
))
TESTS.append((
"Rotated 3d",
"3D",
1e-1,
Rotated(KEYS, [random.uniform(np.pi / 6, np.pi) for _ in range(3)],
True), # type: ignore
))
TESTS.append((
"RandRotated 3d",
"3D",
1e-1,
TESTS.append(("RandZoom 3d", "3D", 9e-2, True,
RandZoomd(KEYS,
1, [0.5, 0.6, 0.9], [1.1, 1, 1.05],
keep_size=True)))
TESTS.append(("RandRotated, prob 0", "2D", 0, True,
RandRotated(KEYS, prob=0, dtype=np.float64)))
TESTS.append((
"Rotated 2d",
"2D",
8e-2,
True,
Rotated(KEYS,
random.uniform(np.pi / 6, np.pi),
keep_size=True,
align_corners=False,
dtype=np.float64),
))
TESTS.append((
"Rotated 3d",
"3D",
1e-1,
True,
Rotated(KEYS, [random.uniform(np.pi / 6, np.pi) for _ in range(3)],
True,
dtype=np.float64),
))
TESTS.append((
