Table 4.
Influence of the channel-attention (channel-att) and multi-grained (Mg) convolution modules on the Human3.6M and CMU-MoCap datasets. On average, the two components of our model contributed to its accuracy. The best results are marked in bold.
| Human3.6M MPJPE (mm) | CMU-MoCap MPJPE (mm) | ||||||||||||
| Channel-att | MgTCN | 80 | 160 | 320 | 400 | 560 | 1000 | 80 | 160 | 320 | 400 | 560 | 1000 |
| √ | 10.4 | 22.9 | 48.1 | 59.0 | 75.9 | 108.7 | 7.7 | 14.5 | 29.0 | 36.1 | 51.9 | 81.7 | |
| √ | 10.5 | 23.1 | 48.0 | 59.4 | 76.8 | 111.0 | 7.9 | 14.6 | 29.2 | 36.7 | 52.3 | 82.6 | |
| √ | √ | 10.2 | 22.4 | 46.9 | 57.7 | 74.8 | 106.7 | 7.5 | 14.0 | 28.1 | 34.8 | 49.0 | 79.4 |