VISUAL GEOMETRY GROUNDED TRANSFORMER

Summary from the TLDR AI newsletter

VGGT is a feed-forward neural network that directly infers all key 3D attributes of a scene, including extrinsic and intrinsic camera parameters, point maps, depth maps, and 3D point tracks, from one, a few, or hundreds of its views, within seconds.