[CVPR'25 Oral] MoGe: Unlocking Accurate Monocular Geometry Estimation for Open-Domain Images with Optimal Training Supervision
[CORL2026] Official implementation of the paper: "OptiGeo: Efficient Monocular Geometry for Embodied Perception in Optically Challenging Scenes"