Description
Two photos with very different viewpoints defeat classic feature matching, stalling 3D reconstruction. MASt3R from NAVER is an image matching model grounded in 3D.
It predicts dense 3D points and local features together, matching precisely under large viewpoint changes for reconstruction and localization.
3D matching:Viewpoint-robust.
Dense:Precise correspondences.
Reconstruction:Multi-view.
Localization:Camera poses.
It predicts dense 3D points and local features together, matching precisely under large viewpoint changes for reconstruction and localization.
Features
3D matching:Viewpoint-robust.
Dense:Precise correspondences.
Reconstruction:Multi-view.
Localization:Camera poses.
