Reproduce “SettleSplat: Monocular Gaussian Mapping under In-View Scene Changes” 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

Anh Thuan Tran, Xiyun Song, Yubin Zhou, Heather Yu, Jana Kosecka, Liang Peng

SettleSplat updates a 3D map as a camera observes moving people and objects that settle in new locations. It represents the scene with small renderable Gaussian elements, assigns them dynamicity using DINOv2-derived tracking uncertainty, and uses uncertainty-weighted depth disagreement to find stale geometry. On Co-Fusion change-region novel views, it averages 33.12 dB PSNR versus DROID-W’s 25.67, across four sequences; tests cover selected in-view changes.

Summary by alphaXiv.

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