Transfer multi-source knowledge via scale-aware online domain adaptation in depth estimation for autonomous driving
Abstract: Highlights•Our meta-learning-based online domain adaptation can exploit multi-source domains.•We introduce a camera conversion technique for normalizing images and depth cues.•Unreliable optimization of inconsistent regions is alleviated by gradient surgery.•Our method achieves virtual-to-real SOTA and can infer metrically accurate depth.
External IDs:dblp:journals/ivc/HuyenBNPDN24
Loading