def test_apply_affine_transform():
image = np.zeros((5, 5, 2))
image[..., 0] = [[0, 0, 1, 0, 0], [0, 1, 0, 1, 0], [1, 0, 0, 0, 1],
[0, 1, 0, 1, 0], [0, 0, 1, 0, 0]]
image[..., 1] = [[1, 1, 1, 1, 1], [0, 0, 1, 0, 0], [1, 1, 1, 1, 1],
[0, 0, 1, 0, 0], [0, 0, 1, 0, 0]]
res = apply_affine_transform(image, theta=90, mode='constant', cval=0)
expected = np.zeros((5, 5, 2))
expected[..., 0] = [[0, 0, 1, 0, 0], [0, 1, 0, 1, 0], [1, 0, 0, 0, 1],
[0, 1, 0, 1, 0], [0, 0, 1, 0, 0]]
expected[..., 1] = [[0, 0, 1, 0, 1], [0, 0, 1, 0, 1], [1, 1, 1, 1, 1],
[0, 0, 1, 0, 1], [0, 0, 1, 0, 1]]
assert np.allclose(res, expected)
image = np.zeros((5, 5, 2))
image[..., 0] = [[0, 0, 1, 0, 0], [0, 1, 0, 1, 0], [1, 0, 0, 0, 1],
[0, 1, 0, 1, 0], [0, 0, 1, 0, 0]]
image[..., 1] = [[1, 1, 1, 1, 1], [0, 0, 1, 0, 0], [1, 1, 1, 1, 1],
[0, 0, 1, 0, 0], [0, 0, 1, 0, 0]]
res = apply_affine_transform(image, shift=(1, 0), mode='constant', cval=0)
expected = np.zeros((5, 5, 2))
expected[..., 0] = [[0, 1, 0, 1, 0], [1, 0, 0, 0, 1], [0, 1, 0, 1, 0],
[0, 0, 1, 0, 0], [0, 0, 0, 0, 0]]
expected[..., 1] = [[0, 0, 1, 0, 0], [1, 1, 1, 1, 1], [0, 0, 1, 0, 0],
[0, 0, 1, 0, 0], [0, 0, 0, 0, 0]]
assert np.allclose(res, expected)
assert np.all(np.rint(res) == expected)
def transform(self, images, targets):
transformed_images = images.copy()
transformed_targets = targets.copy()
# loop through
for i in range(len(images)):
# apply affine transform if possible
if self.affine_transform:
# After affine transform, the pixel intensity may change
# the image should clip back to original range
reduced_ax = tuple(range(len(transformed_images[i].shape) - 1))
vmin = transformed_images[i].min(axis=reduced_ax)
vmax = transformed_images[i].max(axis=reduced_ax)
transformed_images[i] = apply_affine_transform(
transformed_images[i],
mode=self.fill_mode,
cval=self.cval,
theta=self.rotation_degree,
rotation_axis=self.rotation_axis,
zoom_factor=self.zoom_factor,
shift=self.shift).clip(vmin, vmax)
transformed_targets[i] = apply_affine_transform(
transformed_targets[i],
mode=self.fill_mode,
cval=self.cval,
theta=self.rotation_degree,
rotation_axis=self.rotation_axis,
zoom_factor=self.zoom_factor,
shift=self.shift)
# round the target label back to integer
transformed_targets[i] = np.rint(transformed_targets[i])
# flip image
if self.flip_axis is not None:
transformed_images[i] = apply_flip(transformed_images[i],
self.flip_axis)
transformed_targets[i] = apply_flip(transformed_targets[i],
self.flip_axis)
return transformed_images, transformed_targets
def test_apply_affine_transform_3d_correct():
image = np.random.random((4, 4, 4, 2))
res = apply_affine_transform(image,
theta=30,
rotation_axis=0,
shift=[-1, 1, 2],
mode='constant',
cval=0)
expected = apply_affine_transform(image,
theta=30,
rotation_axis=0,
shift=[-1, 1, 2],
mode='constant',
cval=0,
use_3d_transform=True)
assert np.allclose(res, expected)
image = np.random.random((4, 4, 4, 2))
res = apply_affine_transform(image,
theta=-30,
rotation_axis=1,
shift=[2, -1, 1],
mode='constant',
cval=0)
expected = apply_affine_transform(image,
theta=-30,
rotation_axis=1,
shift=[2, -1, 1],
mode='constant',
cval=0,
use_3d_transform=True)
assert np.allclose(res, expected)
image = np.random.random((4, 4, 4, 2))
res = apply_affine_transform(image,
theta=90,
rotation_axis=2,
shift=[1, 2, -1],
mode='constant',
cval=0)
expected = apply_affine_transform(image,
theta=90,
rotation_axis=2,
shift=[1, 2, -1],
mode='constant',
cval=0,
use_3d_transform=True)
assert np.allclose(res, expected)
def test_apply_affine_transform_3d_nosplit():
image = np.zeros((3, 3, 3, 2))
# 3d T in the first channel
image[0][..., 0] = [[0, 0, 0], [0, 1, 0], [0, 0, 0]]
image[1][..., 0] = [[0, 0, 0], [0, 1, 0], [0, 0, 0]]
image[2][..., 0] = [[0, 0, 0], [1, 1, 1], [0, 0, 0]]
# 3d H in the second channel
image[0][..., 1] = [[1, 0, 1], [0, 0, 0], [0, 0, 0]]
image[1][..., 1] = [[1, 1, 1], [0, 0, 0], [0, 0, 0]]
image[2][..., 1] = [[1, 0, 1], [0, 0, 0], [0, 0, 0]]
res = apply_affine_transform(image,
theta=90,
rotation_axis=0,
mode='constant',
cval=0,
use_3d_transform=True)
expected = np.zeros((3, 3, 3, 2))
# 3d T in the first channel
expected[0][..., 0] = [[0, 0, 0], [0, 1, 0], [0, 0, 0]]
expected[1][..., 0] = [[0, 0, 0], [0, 1, 0], [0, 0, 0]]
expected[2][..., 0] = [[0, 1, 0], [0, 1, 0], [0, 1, 0]]
# 3d H in the second channel
expected[0][..., 1] = [[0, 0, 1], [0, 0, 0], [0, 0, 1]]
expected[1][..., 1] = [[0, 0, 1], [0, 0, 1], [0, 0, 1]]
expected[2][..., 1] = [[0, 0, 1], [0, 0, 0], [0, 0, 1]]
assert np.allclose(res, expected)
assert np.all(np.rint(res) == expected)
image = np.zeros((3, 3, 3, 2))
# 3d T in the first channel
image[0][..., 0] = [[0, 0, 0], [0, 1, 0], [0, 0, 0]]
image[1][..., 0] = [[0, 0, 0], [0, 1, 0], [0, 0, 0]]
image[2][..., 0] = [[0, 0, 0], [1, 1, 1], [0, 0, 0]]
# 3d H in the second channel
image[0][..., 1] = [[1, 0, 1], [0, 0, 0], [0, 0, 0]]
image[1][..., 1] = [[1, 1, 1], [0, 0, 0], [0, 0, 0]]
image[2][..., 1] = [[1, 0, 1], [0, 0, 0], [0, 0, 0]]
res = apply_affine_transform(image,
shift=(1, 0, 0),
mode='constant',
cval=0,
use_3d_transform=True)
expected = np.zeros((3, 3, 3, 2))
# 3d T in the first channel
# expected[0][..., 0] = [[0, 0, 0],
# [0, 1, 0],
# [0, 0, 0]]
expected[0][..., 0] = [[0, 0, 0], [0, 1, 0], [0, 0, 0]]
expected[1][..., 0] = [[0, 0, 0], [1, 1, 1], [0, 0, 0]]
# 3d H in the second channel
expected[0][..., 1] = [[1, 1, 1], [0, 0, 0], [0, 0, 0]]
expected[1][..., 1] = [[1, 0, 1], [0, 0, 0], [0, 0, 0]]
assert np.allclose(res, expected)
assert np.all(np.rint(res) == expected)
res = apply_affine_transform(image,
shift=(0, 1, 0),
mode='constant',
cval=0,
use_3d_transform=True)
expected = np.zeros((3, 3, 3, 2))
expected[0][..., 0] = [
[0, 1, 0],
[0, 0, 0],
[0, 0, 0],
]
expected[1][..., 0] = [[0, 1, 0], [0, 0, 0], [0, 0, 0]]
expected[2][..., 0] = [[1, 1, 1], [0, 0, 0], [0, 0, 0]]
assert np.allclose(res, expected)
assert np.all(np.rint(res) == expected)
res = apply_affine_transform(image,
shift=(0, 0, 1),
mode='constant',
cval=0,
use_3d_transform=True)
expected = np.zeros((3, 3, 3, 2))
# 3d T in the first channel
expected[0][..., 0] = [[0, 0, 0], [1, 0, 0], [0, 0, 0]]
expected[1][..., 0] = [[0, 0, 0], [1, 0, 0], [0, 0, 0]]
expected[2][..., 0] = [[0, 0, 0], [1, 1, 0], [0, 0, 0]]
# 3d H in the second channel
expected[0][..., 1] = [[0, 1, 0], [0, 0, 0], [0, 0, 0]]
expected[1][..., 1] = [[1, 1, 0], [0, 0, 0], [0, 0, 0]]
expected[2][..., 1] = [[0, 1, 0], [0, 0, 0], [0, 0, 0]]
assert np.allclose(res, expected)
assert np.all(np.rint(res) == expected)
return normalize(image), target
if __name__ == "__main__":
theta = 30
zoom = 1
rotation_axis = 0
shift = (0, 0, 0)
image, target = load_images()
shape = image.shape[:-1]
transformed = apply_affine_transform(image,
mode='constant',
rotation_axis=rotation_axis,
theta=theta,
zoom_factor=zoom,
shift=shift).clip(0, 1)
transformed_label = (apply_affine_transform(target,
mode='constant',
rotation_axis=rotation_axis,
theta=theta,
zoom_factor=zoom,
shift=shift).clip(0, 1) >
0.5).astype(int)
fig, axes = plt.subplots(2, 2)
for ax in axes.flatten():
ax.axis('off')
nrow = 5
ncol = 8
i = 0
i += 1
plt.subplot(nrow, ncol, i)
plt.axis('off')
plt.imshow(img_slice[..., 0])
plt.title('Original')
i += 1
plt.subplot(nrow, ncol, i)
plt.axis('off')
plt.imshow(img_slice[..., 1])
transformed = apply_affine_transform(img_slice, theta=-16, rotation_axis=2)
transformed[..., 0] = transformed[..., 0].clip(-100, 100)
transformed[..., 1] = transformed[..., 1].clip(0, img_slice[..., 1].max())
i += 1
plt.subplot(nrow, ncol, i)
plt.axis('off')
plt.imshow(transformed[..., 0])
plt.title('Rotate -16')
i += 1
plt.subplot(nrow, ncol, i)
plt.axis('off')
plt.imshow(transformed[..., 1])
def affine_transform():
img = load_image('imgs/cat.jpg')
nrow = 6
ncol = 10
rotate = [0, -30, 15]
zoom = [1, 0.5, 0.8, 1.2, 2]
shift = [(0, 0), (0, 50), (70, 0), (70, 50)]
mode = 'constant'
cval = 0.0
s = 0
i = 1
pos = 1
for row, r in enumerate(rotate):
for col, z in enumerate(zoom):
transformed = apply_affine_transform(img,
theta=r,
zoom_factor=z,
shift=shift[s],
mode=mode, cval=cval)
plt.subplot(nrow, ncol, pos+row*10+col)
plt.axis('off')
plt.imshow(transformed)
plt.title(f'theta={r}, z={z},s={shift[s]}', fontsize=6)
s = 1
i = 1
pos = 6
for row, r in enumerate(rotate):
for col, z in enumerate(zoom):
transformed = apply_affine_transform(img,
theta=r,
zoom_factor=z,
shift=shift[s],
mode=mode, cval=cval)
plt.subplot(nrow, ncol, pos+row*10+col)
plt.axis('off')
plt.imshow(transformed)
plt.title(f'theta={r}, z={z},s={shift[s]}', fontsize=6)
s = 2
i = 1
pos = 31
for row, r in enumerate(rotate):
for col, z in enumerate(zoom):
transformed = apply_affine_transform(img,
theta=r,
zoom_factor=z,
shift=shift[s],
mode=mode, cval=cval)
plt.subplot(nrow, ncol, pos+row*10+col)
plt.axis('off')
plt.imshow(transformed)
plt.title(f'theta={r}, z={z},s={shift[s]}', fontsize=6)
s = 3
i = 1
pos = 36
for row, r in enumerate(rotate):
for col, z in enumerate(zoom):
transformed = apply_affine_transform(img,
theta=r,
zoom_factor=z,
shift=shift[s],
mode=mode, cval=cval)
plt.subplot(nrow, ncol, pos+row*10+col)
plt.axis('off')
plt.imshow(transformed)
plt.title(f'theta={r}, z={z},s={shift[s]}', fontsize=6)
plt.show()