Reproduce “UGG: Unified Generative Geometry Model” 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
Amani Kiruga, Zhiyi Li, Ruojin Cai, Hansen Lillemark, Xiaoshan Wu, Ayush Tewari, Yilun Du
UGG unifies video generation and geometric prediction by completing aligned RGB, depth, and camera-ray representations from observed scene components. A shared multimodal model treats observed tokens as context and denoises missing ones; test-time search ranks candidate completions using image-derived evidence. On RealEstate10K with only endpoint images, search lowers translation-direction angular error from 8.260° to 3.475°, but verification is imperfect.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.