Reproduce “Depth Pro 2: Sharp and Temporally Consistent Metric Monocular Depth” 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

Peiyun Hu, Marko Mihajlovic, Lyndon Duong, Lahav Lipson, Wei Dong, Amaël Delaunoy, Stephan R. Richter, Vladlen Koltun

Depth Pro 2 estimates metric depth from images and video by processing frames jointly at native resolution with within-frame and cross-frame attention. In zero-shot comparisons, H+ ranks first for video metric-depth accuracy, boundary sharpness, and temporal consistency. On Sintel with predicted intrinsics, H+ reaches 41.4% δ1 versus 37.0% for UniDepthV2-L; rankings summarize separate measures, and results vary by model size.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.