Description
Reconstructing 3D space from photos or video traditionally needs camera parameters and complex multi-stage pipelines. Depth Anything 3 from ByteDance Seed recovers visual space from any number of views, down to depth from a single image.
It unifies monocular depth, multi-view reconstruction and camera pose estimation, with models and a project page.
Any views:One to many images.
Depth:Accurate depth maps.
Pose:Camera positions.
3D:Recovered space.
It unifies monocular depth, multi-view reconstruction and camera pose estimation, with models and a project page.
Features
Any views:One to many images.
Depth:Accurate depth maps.
Pose:Camera positions.
3D:Recovered space.
