MoGe

MoGe: Unlocking Accurate Monocular Geometry Estimation for Open-Domain Images with Optimal Training Supervision

Ruicheng Wang et al. · CVPR 2025 Oral

Read paper · arXiv:2410.19115

Use robust geometric alignment to handle scale and depth-shift ambiguity in monocular predictions.

What you will implement

Robust optimal estimation with truncated weighted L1 residuals.

• 47. MoGe Robust Optimal Alignment

  Solve shared scale and z translation exactly using weighted L1 breakpoints and depth anchors.

  Implement the global ROE alignment core on tiny point sets; exclude the point-map network, multi-scale spheres and camera recovery.