Description
Recovering materials and lighting from a few photos usually takes tens of minutes of per-scene optimization with inconsistent results. MVInverse is a feed-forward multi-view inverse rendering framework that estimates consistent geometry and materials in seconds.
No per-scene optimization; inference and training code are open, published at CVPR 2026.
Feed-forward:Seconds.
Consistent:Across views.
Materials:Albedo, roughness and more.
No per-scene optimization; inference and training code are open, published at CVPR 2026.
Features
Feed-forward:Seconds.
Consistent:Across views.
Materials:Albedo, roughness and more.
