Reproduce “Chat-Scene++: Exploiting Context-Rich Object Identification for 3D LLM” 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
Haifeng Huang, Yilun Chen, Zehan Wang, Jiangmiao Pang, Zhou Zhao
Chat-Scene++ represents scenes as sequences of object proposals paired with identifiers and context-aware three-dimensional and multi-view two-dimensional features, enabling language models to refer to objects and reason across tasks. With Vicuna-7B and two-dimensional input, it scores 58.7 [email protected] on Multi3DRefer, versus Chat-Scene’s 52.4; however, 2D-only ScanNet grounding remains limited by video-tracking errors.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.