Reproduce “RelationVGGT: Visual Geometry Transformers for 3D Spatial Relation Segmentation” 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
Minsu Kim, Jaesung Choe, Jiwoo Lee, Yu-Chiang Frank Wang, Seon Joo Kim
RelationVGGT finds cross-view relational masks from unposed multi-view images, given a subject mask and relation phrase. It combines DINOv2 and Pi3 features, then uses subject-conditioned attention to predict relation features on target-view pixel grids rather than searching a reconstructed 3D volume. On three benchmarks it achieves macro mIoU 0.382 and mAcc 0.549, but trails RelationField on cluttered LERF scenes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.