Reproduce “BEV3SR: Evidence-Aware 3D Spatial Reasoning from Video” 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
Fan Zhu, Rui Yu
BEV3SR uses RGB-D video and camera parameters to build persistent metric scene memory linking object locations to source views. It formalizes questions and routes them to geometric computation or targeted visual retrieval. On SQA3D test with Qwen3-VL-8B, exact-match accuracy rises from 51.6 to 62.6 versus uniform 16-frame RGB prompting with the same backbone and frame budget, though input modalities differ.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.