Self-Supervised Learning of Depth and Ego-Motion From Videos by Alternative Training and Geometric Constraints from 3-D to 2-D | IEEE Journals & Magazine | IEEE Xplore