An unofficial pytorch implement of HCN and Two-stream CNN for skeleton-based action recognition.
It will be updated soon......
| model | cross view | cross subject | gpu |
|---|---|---|---|
| Two Stream CNN | 89.3% | 83.2% | |
| Two stream CNN (re) | 89.1% | 83.7% | gtx 1070 |
| HCN | 91.1% | 86.5% | |
| HCN (re) | 91.5% | 86.4% | gtx 1070 |
| HCN (re) | 91.6% | 86.5% | gtx 1080ti |
| HCN-P | 88.8% | 82.4% | gtx 1070 |
| HCN-M | 87.8% | 82.0% | gtx 1070 |