Reproduce “Kosmo-GFM: A Production-Ready Geometry Foundation 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
Sheng Tang, Shengzhe Wei, Tiancheng Chen, Wenhua Jin, Xinyang Lu, Tinghao Jiang, Shuoxuan Feng, Yupeng Liu and 24 more
Kosmo-GFM combines geometric predictions with optimization to reconstruct cameras and geometry from photos and video. Its PoseEngine system uses foundation-model priors and cross-view constraints, while KosmoWorld provides training examples from real-scene reconstructions. On 34 completed scenes, PoseEngine achieved AUC@5 of 94.26% versus GLUEMAP’s 90.66%, with runtimes of 14.07 versus 69.67 hours; this system comparison does not attribute gains to GFM alone.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.