Pose-Only Supervised Sparse Visual Odometry with Salient Patch Detection and Self-Supervised Homographic Pre-Training

This paper presents a novel hybrid visual odometry framework that leverages pose-only supervision, offering a balanced solution between robustness and the need for extensive labeling. It introduces a self-supervised homographic pre-training method for enhancing optical flow learning from pose-only labels and a random patch-based salient point detection strategy for more accurate optical flow patch extraction.