Dyna-DM: Dynamic Object-aware Self-supervised Monocular Depth Maps

TL;DR

This paper introduces Dyna-DM, a self-supervised monocular depth estimation method that improves performance by focusing on dynamic objects and refining the learning process, reducing memory usage and enhancing depth map quality.

Contribution

It proposes a novel approach that emphasizes learning improvements over model complexity, including ignoring small dynamic objects and separately estimating object pose for truly dynamic objects.

Findings

Reduces GPU memory usage by 29%

Achieves improved depth map quality

Demonstrates effectiveness on benchmark datasets

Abstract

Self-supervised monocular depth estimation has been a subject of intense study in recent years, because of its applications in robotics and autonomous driving. Much of the recent work focuses on improving depth estimation by increasing architecture complexity. This paper shows that state-of-the-art performance can also be achieved by improving the learning process rather than increasing model complexity. More specifically, we propose (i) disregarding small potentially dynamic objects when training, and (ii) employing an appearance-based approach to separately estimate object pose for truly dynamic objects. We demonstrate that these simplifications reduce GPU memory usage by 29% and result in qualitatively and quantitatively improved depth maps. The code is available at https://github.com/kieran514/Dyna-DM.