Metric3Dv2: A Versatile Monocular Geometric Foundation Model for Zero-shot Metric Depth and Surface Normal Estimation

Benchmark Model Rank Results
monocular-depth-estimation-on-ibims-1Metric3D-v2(L, ZS)#4δ1.25: 0.969
monocular-depth-estimation-on-kitti-eigenMetric3Dv2 (g2, FT, 80m, flip_aug_test)#4absolute relative error: 0.039RMSE: 1.766RMSE log: 0.060
monocular-depth-estimation-on-nyu-depth-v2Metric3Dv2(L, FT)#6absolute relative error: 0.047RMSE: 0.183log 10: 0.020
surface-normals-estimation-on-nyu-depth-v2-1Metric3Dv2(L, FT)#1% < 11.25: 68.8% < 22.5: 84.9% < 30: 89.8Mean Angle Error: 12.0
surface-normals-estimation-on-scannetv2Metric3Dv2 (g2, In-domain)#1% < 11.25: 77.8% < 22.5: 90.1% < 30: 93.5Mean Angle Error: 9.2