Peking University Team Open-Sources SLAM3R, Rebuilding 3D Scenes From a Single Video at 20 FPS

Researchers at Peking University have released SLAM3R, an open-source system that reconstructs dense 3D scenes in real time from ordinary monocular video—no LiDAR, depth sensors or specialized rigs required. Using a single camera, the system rebuilds an entire scene in 3D at more than 20 frames per second on a single GPU, and reportedly remains stable across long sequences of over 10,000 frames without collapsing.