Table 6 Performance comparisons on NTU60 with the CS and CV settings in top-1 accuracy.
From: An efficient self-attention network for skeleton-based action recognition
Methods | Parameters (M) | cs (%) | cv (%) |
|---|---|---|---|
GCA-LSTM28 | – | 74.4 | 82.8 |
VA-LSTM29 | – | 79.2 | 87.7 |
ST-GCN14 | – | 81.5 | 88.3 |
DPRL+GCNN30 | – | 83.5 | 89.8 |
SR-TSL31 | – | 84.8 | 92.4 |
AS-GCN18 | 7.40 | 86.8 | 94.2 |
GR-GCN17 | – | 84.8 | 92.4 |
2s-AGCN7 | 6.92 | 88.5 | 95.1 |
AGC-LSTM15 | 22.81 | 89.2 | 95.0 |
2s-SDGCN13 | – | 89.6 | 95.7 |
SGN8 | 0.69 | 89.0 | 94.5 |
DGNN32 | 8.16 | 89.9 | 96.1 |
Shift-GCN (2s)10 | 1.48 | 89.7 | 96.0 |
Shift-GCN (4s)10 | 2.94 | 90.7 | 96.5 |
MS-G3D (joint)11 | 3.20 | 89.4 | 95.0 |
MS-G3D (2s)11 | 6.40 | 91.5 | 96.2 |
MST (joint)33 | 3.0 | 89.0 | 95.1 |
MST (2s)33 | 6.0 | 91.1 | 96.4 |
Double-head (joint)34 | 3.0 | 90.3 | 96.1 |
Double-head (2s)34 | 6.0 | 91.7 | 96.5 |
Ours | 0.89 | 90.5 | 96.1 |
