def execute(self, image, mask):
x, y, z, _ = image.shape
image = image.reshape((x, y, z))
mask = np.argmax(mask, axis=-1)
newx = x + int(scale_truncated_norm(self.std))
newy = y + int(scale_truncated_norm(self.std))
newz = z + int(scale_truncated_norm(self.std))
image = scale3d(image, newx, newy, newz, cv2.INTER_CUBIC)
mask = scale3d(mask, newx, newy, newz, cv2.INTER_NEAREST)
image = image.reshape(image.shape + (1,))
mask = to_channels(mask)
return image, mask
def execute(self, image, mask):
x, y, z, _ = image.shape
dx, dy, dz = abs(int(scale_truncated_norm(self.std))), abs(
int(scale_truncated_norm(self.std))), abs(int(scale_truncated_norm(self.std)))
newx = x - dx
newy = y - dy
newz = z - dz
sx, sy, sz = np.random.choice(range(dx)), np.random.choice(
range(dy)), np.random.choice(range(dz))
image = image[sx: sx + newx, sy: sy + newy, sz: sz + newz, :]
mask = mask[sx: sx + newx, sy: sy + newy, sz: sz + newz, :]
return image, mask
def random_affine_warp_fn(self, vertex_percentage_std):
source_points = [[0, 0, 0], [0, 100, 0], [0, 0, 100], [100, 0, 0]]
dest_points = np.array(
[[v + scale_truncated_norm(vertex_percentage_std)[0] for v in p]
for p in source_points])
return lambda img, is_mask: self.affine_warp(img, source_points,
dest_points, is_mask)
def apply_variability(value, std):
res = value + scale_truncated_norm(std)[0]
if res > 1:
res = 1
if res < 0:
res = 0
return res
def random_elastic_deform_fn(self, sigma_std, possible_points):
sigma = scale_truncated_norm(sigma_std)
points = np.random.choice(possible_points)
return lambda img, mask: self.elastic_deform(img, mask, sigma, points)
def random_swirl_fn(self, strength_std, r):
ax = random.choice(AXES)
s = scale_truncated_norm(strength_std)
return lambda img, is_mask: self.swirl(img, ax, s, r, is_mask)
def random_shift_fn(self, shift_stds):
shifts = [scale_truncated_norm(s) for s in shift_stds]
return lambda img, is_mask: self.shift(img, shifts, is_mask)
def random_uniaxial_rotation_fn(self, std):
d = scale_truncated_norm(std)
a = random.choice(AXES)
return lambda img, is_mask: self.rotate(img, d, a, is_mask)
def random_linear_gradient_fn(stds):
gx, gy, gz = [scale_truncated_norm(s)[0] for s in stds]
return lambda img, is_mask: apply_gradient(
img, lambda x, y, z: gx * x + gy * y + gz * z, is_mask)
