why do the "line[-1] = angle_offset +np.arctan(float(line[11]) / float(line[13]))" ??

[SeoJaeMin]

I don't know why you using the kitti GT annotation during inference??
That term "line[11], line[13]" are from kitti GT annotation.

[yingsen1]

i have the same question, and i wonder: we caculate translation T through yaw angle. But we obtain yaw through local_yaw (by network) + arctan(Tx/Tz). i am confused about that. or am i wrong?

[marshallixp]

In the test function of main.py, the predicted value is not used for evaluation. So, the result is confused and doubted.

[LC041336]

Maybe the original paper is the best answer. The paper shows that "theta = theta_ray + theta_l".theta is the yaw of the vehicle, theta_ray is arctan(X_camera_coordinate / Z_camera_coordinate),theta_l is the local orientation.
So the angle_offset in the code refers to the theta_l, which is the result of CNN.