Horizon Releases Video Geometry Base Model LoFT
Horizon Robotics Laboratory released the native video geometry base model LoFT on November 18, which can jointly predict camera pose, intrinsics, and per-frame depth from RGB video to recover the 3D structure of the scene. The model adopts a dual-loop Transformer architecture, leading in multiple comprehensive evaluations, and demonstrates geometric reconstruction and cross-scene generalization capabilities in scenarios such as robotic manipulation, first-person interaction, and environmental exploration.