Table 4 Performance changes of the model on different data sets after removing different modules.
From: Basketball technique action recognition using 3D convolutional neural networks
Experiment | 3D convolutional layer | LSTM layer | Pooling layer | Fully connected layer | Accuracy on NTURGB + D | Accuracy on the Basketball-Action-Dataset | Accuracy on the B3D Dataset |
|---|---|---|---|---|---|---|---|
1 | Enabled | Enabled | Enabled | Enabled | 93.1% | 88.5% | 90.0% |
2 | Disabled | Enabled | Enabled | Enabled | 85.2% | 80.1% | 82.3% |
3 | Enabled | Disabled | Enabled | Enabled | 89.1% | 84.2% | 85.7% |
4 | Enabled | Enabled | Disabled | Enabled | 90.5% | 85.8% | 87.2% |
5 | Enabled | Enabled | Enabled | Disabled | 79.3% | 75.6% | 78.1% |
