Deep Visual Odometry with Adaptive Memory

Fei Xue, Xin Wang, Junqiu Wang +1

We propose a novel deep visual odometry (VO) method that considers global information by selecting memory and refining poses. Existing learning-based methods take the VO task as a…