Summary
PURE reconstructs depth, camera poses, and colored 3D point clouds from video clips or image sequences. Its unlabeled post-training approach improves feed-forward reconstruction backbones including Pi3, Depth Anything 3, and VGGT.
PURE reconstructs depth, camera poses, and colored 3D point clouds from video clips or image sequences. Its unlabeled post-training approach improves feed-forward reconstruction backbones including Pi3, Depth Anything 3, and VGGT.
Researchers and creators can compare reconstruction backbones and inspect the resulting geometry through an interactive studio and released models. Core capabilities include Video-to-3D reconstruction, Image-sequence input, Depth estimation, and Camera-pose recovery.
Public code or model weights are linked from the project, with setup instructions for running or adapting the system. Additional characteristics include Colored point-cloud output, Pi3 backbone support, DA3 and VGGT support, and Interactive PURE Studio.
Get more likes & reach the top of search results by adding this button on your site!


Get top updates in AI to your inbox every weekend. It's free!