Reproduce “Teaching DINOv3 About Partial 3D Geometry: A Self-Supervised Geometry-Aware Approach” 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
Viktoria Ehm, Dongliang Cao, Riccardo Marin, Daniel Scholz, Weikang Wang, Florian Bernard, Daniel Cremers
GeoLoRA adapts a frozen DINOv3 image model to match partial and complete 3D shapes, training LoRA adapters on synthetic partial shapes with known correspondences and a geodesic-aware contrastive loss. On BECOS partial-to-full matching, its mean geodesic error is 5.57, compared with 19.95 for DINOv3. Real-scan evidence remains illustrative, not a broad benchmark.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.