Reproduce “Lie Routing Vision Language Models for View-Robust 3D Understanding” 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

Jingjun Yi, Chen Hu, Qi Bi, Hao Zheng, Huimin Huang, Yongkang, Wei Ji, Yawen Huang and 2 more

LieVLM addresses viewpoint sensitivity in 3D vision-language tasks by using geometry features to steer a structured correction to visual features while retaining visual and geometry paths. On ScanRefer, it reaches 64.56 ± 0.31 [email protected], versus 61.27 ± 0.25 for a rational-controlled generator in a three-seed ablation. This supports the combined design, but does not isolate rotation structure as the cause or establish viewpoint invariance.

Summary by alphaXiv.

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