Research Article
Lite-3DCNN Combined with Attention Mechanism for Complex Human Movement Recognition
Table 2
The validation accuracy of the proposed method for the complex human movement of the UCF-101 dataset.
| Model | 10 class (%) | 20 class (%) | 30 class (%) |
| C3D | 82.2 | 84.7 | 83.5 | Lite-3DCNN | 85.3 | 80.2 | 70.6 | Lite-3DCNN-LSTM | 81.1 | 83.5 | 75.2 | Lite-3DCNN-BiLSTM | 84.5 | 85.3 | 79.5 | Proposed method | 91.6 | 88.5 | 84.8 |
|
|