def color_coding(flow, rad_org):
max_rad_glo = fl2.compute_maxrad(flow)
if abs(rad_org - max_rad_glo) < 1:
img_ = fl.flow_to_image(flow)
else:
img_ = fl2.flow_to_image(flow, rad_org)
return img_
def reconstruct(data, maxrad, w=200, h=200):
x_data = data[0]
y_data = data[1]
line = np.arange(0, w)
data_com_x = x_data[0] * line.reshape(1, -1)**2 + x_data[1] * line.reshape(
1, -1) + x_data[2]
x_mat = np.repeat(data_com_x, h, axis=0)
col = np.arange(0, h)
data_com_y = y_data[0] * col.reshape(-1, 1)**2 + y_data[1] * col.reshape(
-1, 1) + y_data[2]
y_mat = np.repeat(data_com_y, w, axis=1)
flo_x = x_mat[:, :, np.newaxis]
flo_y = y_mat[:, :, np.newaxis]
flo_global = np.concatenate((flo_x, flo_y), axis=2)
flo_global_color = fl2.flow_to_image(flo_global, maxrad)
plt.imshow(flo_global_color)
plt.show()
interpolation=cv2.INTER_LINEAR)
global_motion = global_motion_estimation(data,
w=data_org.shape[1],
h=data_org.shape[0])
end = time.time()
print('time: ', end - start)
epe = fl.flow_error(global_motion[0], global_motion[1], data_org[0],
data_org[1])
print('epe: {}'.format(epe))
local_motion = data_org - global_motion
local_motion_outlier = local_motion * outlier_set
flo_global_color = fl2.flow_to_image(global_motion, maxrad)
flo_local_color = fl2.flow_to_image(local_motion, maxrad)
local_motion_outlier_rad = np.sqrt(local_motion_outlier[:, :, 0]**2 +
local_motion_outlier[:, :, 1]**2)
local_motion_outlier_mask = np.where(
np.abs(local_motion_outlier_rad) > 1.0, 1, 0)
local_motion_outlier[:, :, 0] *= local_motion_outlier_mask
local_motion_outlier[:, :, 1] *= local_motion_outlier_mask
flo_local_outlier_color = fl2.flow_to_image(local_motion_outlier, maxrad)
fl.visualize_flow(data_org)
plt.imshow(flo_global_color)