VINO sets new ScanNet SOTA for 3D instance segmentation at AP50 83.4
NielsRogge · x · 2026-11-08
- VINO (Revisiting 3D Representation Learning with Volume Transformers) tops the Papers with Code ScanNet leaderboard for 3D instance segmentation with AP50 83.4, beating Volt-B + SPFormer (82.7), OneFormer3D (80.1), and Mask3D (78.0).
- The method brings DINO-style representation learning to 3D/volumetric data; a collaboration between RWTH, USI, and TU Eindhoven, with paper and code publicly available.
- HF's Niels Rogge congratulated the authors on the new state of the art.
More from Research
- AI helps neurosurgery resident settle 22-year-old Crouzeix's conjecture in matrix analysis — IgorCarron · 2026-11-08
- Scale invariance may fix why autoregressive image generation lags diffusion models — burkov · 2026-11-08
- Agent Lightning paper trains models through the same harness used at deployment — bibryam · 2026-11-08
- Radziwill says the quasi-Riemann proof fits in a tweet from the future — Singularitarian · 2026-11-08
- DeepMind exec says scientific discovery is undergoing a 'genuine phase transition' in pace — pushmeet · 2026-11-08
- OpenAI Researcher: Downstream Capability Scales Roughly Linearly, Not Log — willdepue · 2026-11-08