Reproduce “SplatScout: Query-Driven Part-Level Referring Segmentation in Gaussian Scenes” 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
Yihan Zang, Da Li, Shinkyu Park, Ivan Viola, Dominik Engel
SplatScout lets a vision-language model direct segmentation of queried regions in reconstructed 3D scenes. Instead of choosing from fixed candidates, it inspects rendered results and can request revised views or cuts. In a trash-bin example, it corrects an initial lid lift that includes the bag. On PartRef-ScanNet++, it scores 0.538 mIoU versus GaussDet’s 0.253, though methods use different models and candidates.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.