Reproduce “DACA: Dual-Anchor Collaborative Agent for Zero-Shot 3D Visual Grounding” 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

Jiaming Liu, Linghe Kong, Guixu Zhang, Huaxi Huang

DACA develops RGB-D visual grounding by letting appearance tracking and 3D geometry exchange evidence in both directions. It uses optical flow, a vision-language model, and geometric view retrieval to extend object masks across an image sequence and reconstruct a 3D box. On 250 ScanRefer queries, it scores 73.6% [email protected] versus 71.2% for TAB with the same model; at [email protected], scores are 51.2% and 46.4%. These subset results do not establish reliable correction of wrong initial identities.

Summary by alphaXiv.

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