Reproduce “Adapting Monocular Depth Estimation in Novel Scenes through Disparity Refinement and Robot Ego-Motion” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Dimitrios Chatziparaschis, Georgios Georgakis, Adnan I. Ansar, Shehryar Khattak, Konstantinos Karydis
EM-ViT adapts a frozen single-image depth prior from motion to produce metric depth. Its network refines disparity and predicts an effective-baseline scalar, with sparse LiDAR supervision and photometric cues. Across Garage, CitrusFarm, and KITTI, it reduces L1_10 error by 31 cm versus online-adapted Depth Anything V2 and uses 16% less total training time, but does not win every metric.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.