3D Pose Forecasting (Joint Angles) on Human3.6M
0.2MAE @ 80msMotionMixer
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| MotionMixerEvaluation protocol=average error over all frames2022.07 | 0.2 | 0.34 | 0.55 | 0.63 | 0.78 | 0.91 | 0.99 | 1.08 | |
| STS-GCNobserved frames=102021.10 | 0.24 | 0.39 | 0.59 | 0.66 | 0.79 | 0.92 | 1 | 1.09 | |
| STSGCNEvaluation protocol=average error over all frames2022.07 | 0.24 | 0.39 | 0.59 | 0.66 | 0.79 | 0.92 | 1 | 1.09 | |
| GAGCNEvaluation protocol=average error over all frames2022.07 | 0.24 | 0.38 | 0.54 | 0.65 | 0.74 | — | — | 1.02 | |
| Motion-AttentionEvaluation protocol=particular frame2022.07 | 0.27 | 0.51 | 0.81 | 0.93 | 1.12 | 1.27 | 1.46 | 1.57 | |
| MotionMixerEvaluation protocol=particular frame2022.07 | 0.29 | 0.54 | 0.81 | 0.94 | 1.2 | 1.3 | 1.4 | 1.57 | |
| BC-WGAIL-divobserved frames=102021.10 | 0.31 | 0.57 | 0.9 | 1.02 | 1.23 | — | — | 1.65 | |
| DCT-RNN-GCNobserved frames=102021.10 | 0.31 | 0.55 | 0.9 | 1.04 | 1.25 | 1.42 | 1.56 | 1.65 | |
| RNN-GCNEvaluation protocol=particular frame2022.07 | 0.31 | 0.55 | 0.9 | 1.04 | 1.25 | 1.42 | 1.56 | 1.65 | |
| LTD-10-10observed frames=10, predicted frames=102021.10 | 0.32 | 0.55 | 0.91 | 1.04 | 1.26 | 1.44 | 1.59 | 1.68 | |
| LTD-10-25Evaluation protocol=particular frame2022.07 | 0.32 | 0.55 | 0.91 | 1.04 | 1.26 | 1.44 | 1.59 | 1.68 | |
| LTD-10-25observed frames=10, predicted frames=252021.10 | 0.34 | 0.57 | 0.93 | 1.06 | 1.27 | 1.44 | 1.57 | 1.66 | |
| Res. sup.Evaluation protocol=particular frame2022.07 | 0.36 | 0.67 | 1.02 | 1.15 | — | — | — | — | |
| convSeq2SeqEvaluation protocol=particular frame2022.07 | 0.38 | 0.68 | 1.01 | 1.13 | 1.35 | 1.5 | 1.69 | 1.82 | |
| MHUEvaluation protocol=particular frame2022.07 | 0.39 | 0.68 | 1.01 | 1.13 | 1.14 | 1.28 | 1.46 | 1.57 |