| 前置网络 | 池化类型 | 准确度 |
| SF Net | Flatten | 86.9% |
| Attention pooling | 93.6% | |
| Our attention pooling | 95.3% | |
| Our attention pooling + SE Block | 95.9% | |
| Cnn_P | Flatten | 93.4% |
| Attention pooling | 93.8% | |
| Our attention pooling | 95.0% | |
| Cnn_NP | Flatten | 88.3% |
| Attention pooling | 91.0% | |
| Our attention pooling | 94.5% |