Proposing a self-supervised method to jointly learn 3D motion and depth from monocular videos, addressing the limitations of existing methods.
