def test_rand_weighted_crop_default_roi(self):
img = self.imt[0]
n_samples = 3
crop = RandWeightedCrop((10, -1, -1), n_samples)
weight = np.zeros_like(img)
weight[0, 7, 17] = 1.1
weight[0, 13, 31] = 1.1
weight[0, 24, 21] = 1
crop.set_random_state(10)
result = crop(img, weight)
self.assertTrue(len(result) == n_samples)
np.testing.assert_allclose(result[0].shape, (1, 10, 64, 80))
np.testing.assert_allclose(np.asarray(crop.centers), [[14, 32, 40], [41, 32, 40], [20, 32, 40]])
def test_rand_weighted_crop_small_roi(self):
img = self.seg1[0]
n_samples = 3
crop = RandWeightedCrop((8, 10, 12), n_samples)
weight = np.zeros_like(img)
weight[0, 5, 30, 17] = 1.1
weight[0, 8, 40, 31] = 1
weight[0, 11, 23, 21] = 1
crop.set_random_state(10)
result = crop(img, weight)
self.assertTrue(len(result) == n_samples)
np.testing.assert_allclose(result[0].shape, (1, 8, 10, 12))
np.testing.assert_allclose(np.asarray(crop.centers), [[11, 23, 21], [5, 30, 17], [8, 40, 31]])
def test_rand_weighted_crop_bad_w(self):
img = self.imt[0]
n_samples = 3
crop = RandWeightedCrop((20, 40), n_samples)
weight = np.zeros_like(img)
weight[0, 30, 17] = np.inf
weight[0, 10, 1] = -np.inf
weight[0, 10, 20] = -np.nan
crop.set_random_state(10)
result = crop(img, weight)
self.assertTrue(len(result) == n_samples)
np.testing.assert_allclose(result[0].shape, (1, 20, 40))
np.testing.assert_allclose(np.asarray(crop.centers), [[63, 37], [31, 43], [66, 20]])
def test_rand_weighted_crop_large_roi(self):
img = self.segn[0]
n_samples = 3
crop = RandWeightedCrop((10000, 400), n_samples)
weight = np.zeros_like(img)
weight[0, 30, 17] = 1.1
weight[0, 10, 1] = 1
crop.set_random_state(10)
result = crop(img, weight)
self.assertTrue(len(result) == n_samples)
np.testing.assert_allclose(result[0].shape, (1, 128, 64))
np.testing.assert_allclose(np.asarray(crop.centers), [[64, 32], [64, 32], [64, 32]])
for res in result:
np.testing.assert_allclose(res, self.segn[0])
def test_rand_weighted_crop(self, _, input_params, img, weight,
expected_shape, expected_vals):
crop = RandWeightedCrop(**input_params)
crop.set_random_state(10)
result = crop(img, weight)
self.assertTrue(len(result) == input_params["num_samples"])
assert_allclose(result[0].shape, expected_shape)
for c, e in zip(crop.centers, expected_vals):
assert_allclose(c, e, type_test=False)
# if desired ROI is larger than image, check image is unchanged
if all(s >= i
for i, s in zip(img.shape[1:], input_params["spatial_size"])):
for res in result:
assert_allclose(res, img, type_test="tensor")
self.assertEqual(len(res.applied_operations), 1)