3D Human Keypoint Estimation on Waymo Open Dataset (WOD) 10% (train)
13.4MPJPE (cm)GC-KPL
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GC-KPLBackbone=Pre-trained on 200,000 WOD-train, Stage I supervised=true2023.06 | 13.4 | -2.96 | |
| GC-KPLBackbone=Pre-trained on 200,000 WOD-train, Stage I supervised=false2023.06 | 14.3 | -2.06 | |
| GC-KPLBackbone=Pre-trained on 5,000 WOD-train, Stage I supervised=false2023.06 | 14.51 | -1.85 | |
| GC-KPLBackbone=Pre-trained on synthetic only, Stage I supervised=false2023.06 | 15.1 | -1.26 | |
| HUM3DILBackbone=Randomly initialized, Stage I supervised=false2023.06 | 16.36 | — |