Reproduce “UniGS-SLAM: Monocular Gaussian SLAM with Enhanced Geometry and Open-Vocabulary Semantics” 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

Xingyuan Ding, Sicheng Yu, Kaiyong Zhao, Hao Wang

UniGS-SLAM keeps geometry in a 3D Gaussian map while storing open-vocabulary semantics in sparse voxels. It retracts and reinserts observations when geometry changes, and uses voxel semantics to guide Gaussian refinement. Across six ScanNet scenes, this revision raises mean class accuracy from 55.85 to 60.80 and mean intersection over union from 45.43 to 49.86; the ablation does not establish retraction as the sole source of full-system improvement.

Summary by alphaXiv.

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