MEOW reconstructs 3D from mixed perspective, fisheye and panorama cameras in a single pass
zhenjun_zhao · x · 2026-09-30
A new paper presents MEOW, a feed-forward 3D reconstruction system that jointly produces metric pointmaps and camera poses from a single N-view tuple mixing perspective, fisheye and 360° panorama images — with no calibration, distortion parameters, camera-type labels, or poses supplied for any view.
Key points:
- MEOW treats heterogeneous-camera reconstruction as a data-adaptation problem rather than an architecture redesign: it keeps a perspective-pretrained backbone and learns heterogeneous cameras entirely from a procedural data engine that renders scenes across a continuous manifold of camera models with exact rays, depth, and certified covisibility.
- Trained only on synthetic tuples, MEOW transfers zero-shot to real captures, scoring 79.9 mAA@30 on heterogeneous 2D3DS tuples vs 53.8 for Wid3R.
More from Research
- Embedding arithmetic demo: cartoon − animation − comic = diagram — mayfer · 2026-10-23
- Omnii Genome: Single Generalist Model Outperforms AlphaGenome Across All Benchmarks — JosephJacks_ · 2026-10-23
- Domingos Cites New Result Claiming RL Is 'Basically Useless' vs Supervised Learning — pmddomingos · 2026-10-23
- Word Arithmetic on Embeddings: "Bathroom Minus Toilet Equals Kitchen" Beats SAE Probing — mayfer · 2026-10-23
- Wolfram Opens Rule 30 Prizes to AIs: $30,000 for Three Open Problems — cephaloform · 2026-10-23
- Stanford and Databricks use AI agents to reproduce medical findings and test analytical robustness — ChengleiSi · 2026-10-23