Description
Reconstructing a scene from photos with COLMAP takes tens of minutes and may fail. VGGT, the CVPR 2025 best paper, is a visual geometry Transformer inferring cameras, depth and point clouds from images in seconds.
One forward pass yields every 3D attribute, far faster than optimization pipelines, with a Hugging Face demo.
Seconds:One pass.
Everything:Cameras, depth and points.
Views:One to hundreds.
Demo:Hugging Face.
One forward pass yields every 3D attribute, far faster than optimization pipelines, with a Hugging Face demo.
Features
Seconds:One pass.
Everything:Cameras, depth and points.
Views:One to hundreds.
Demo:Hugging Face.
