Root joint localization on Human3.6M (test)
77.6MRPEHDNet
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| HDNetBackbone=Volumetric-based 3D pose estimator [29], Input Resolution=256x256, Detector=Mask R-CNN [8]2020.07 | 77.6 | 15.6 | 13.6 | 69.9 | |
| RootNet2020.07 | 120 | 23.3 | 23 | 108.1 | |
| Baseline with RANSACApproach=Two-stage (2D pose [29] + 3D pose [18]), Optimization=Randomly sampling joints with RANSAC2020.07 | 213.1 | 24.3 | 24.3 | 207.1 | |
| Baseline w/o limb jointsApproach=Two-stage (2D pose [29] + 3D pose [18]), Optimization=Only head and body trunk joints2020.07 | 226.2 | 24.5 | 24.9 | 220.2 | |
| BaselineApproach=Two-stage (2D pose [29] + 3D pose [18]), Optimization=Full joints2020.07 | 267.8 | 27.5 | 28.3 | 261.9 |