Unsupervised Monocular Depth and Ego-motion Learning with Structure and Semantics
This work addresses dynamic scene challenges in robotics and autonomous driving, but it is incremental as an extended version of prior research.
The paper tackles unsupervised monocular depth and ego-motion learning by modeling individual object motion jointly with depth and ego-motion, achieving more accurate results, especially in dynamic scenes.
We present an approach which takes advantage of both structure and semantics for unsupervised monocular learning of depth and ego-motion. More specifically, we model the motion of individual objects and learn their 3D motion vector jointly with depth and ego-motion. We obtain more accurate results, especially for challenging dynamic scenes not addressed by previous approaches. This is an extended version of Casser et al. [AAAI'19]. Code and models have been open sourced at https://sites.google.com/corp/view/struct2depth.