Table 7 Performance comparisons on NTU120 with the C-subjects and C-settings in top-1 accuracy.
From: An efficient self-attention network for skeleton-based action recognition
Methods | Year | c-sub (%) | c-set (%) |
|---|---|---|---|
ST-LSTM35 | 2016 | 55.7 | 57.9 |
GCA-LSTM28 | 2017 | 58.3 | 59.2 |
Pose Evolution Map36 | 2018 | 64.6 | 66.9 |
2s-AGCN7 | 2019 | 82.5 | 84.9 |
Shift-GCN10 | 2020 | 85.9 | 87.6 |
MS-G3D11 | 2020 | 86.9 | 88.4 |
SGN8 | 2020 | 79.2 | 81.5 |
MST (joint)33 | 2021 | 82.8 | 84.5 |
MST (2s)33 | 2021 | 87.0 | 88.3 |
Double-head (joint)34 | 2021 | 84.6 | 85.9 |
Double-head (2s)34 | 2021 | 87.9 | 89.1 |
Ours | – | 85.7 | 86.8 |
