Reproduce “MVDG: Efficient Multi-view 3D Disambiguation on Unconstrained Real-World Images” 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
Hanyuan Xiao, Gonglin Chen, Haolin Xiong, Wenbin Teng, Haiwei Chen, Yajie Zhao
MVDG groups images jointly to identify views that share visible surfaces, using a frozen VGGT encoder, layerwise pair classification, and confidence-based early exit. On VisymScenes, it classifies 0.9441 accuracy across 3,180 pairs, versus 0.9355 for Doppelgangers++. Multi-view supervision improves AUC@5 and AUC@10 and SfM inlier ratio, but AUC@3 falls slightly; results are not uniform across benchmarks and scenes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.