Research Article
MAM: Multiple Attention Mechanism Neural Networks for Cross-Age Face Recognition
Table 4
The ACC of the improved model with hierarchical residual-like connections in different deployment modes on the CACD-VS and MORPH datasets.
| | Model | ACC on CACD-VS | ACC on MORPH |
| | ResNet50(fine-tuned by MORPH) | 95.31% | 95.83% | | Residual attention ResNet50 (stage 2-4) | 98.12% | 98.23% | | Residual attention ResNet50 (stage 2-4) with HRC (stage 2) | 98.14% | 98.26% | | Residual attention ResNet50 (stage 2-4) with HRC (stage 3) | 98.17% | 98.28% | | Residual attention ResNet50 (stage 2-4) with HRC (stage 4) | 98.21% | 98.31% | | Residual attention ResNet50 (stage 2-4) with HRC (stage 5) | 98.25% | 98.35% | | Residual attention ResNet50 (stage 2-4) with HRC (stage 2, 3) | 98.24% | 98.43% | | Residual attention ResNet50 (stage 2-4) with HRC (stage 2, 4) | 98.26% | 98.34% | | Residual attention ResNet50 (stage 2-4) with HRC (stage 2-4) | 98.29% | 98.37% | | Residual attention ResNet50 (stage 2-4) with HRC (stage 2-5) | 98.31% | 98.42% | | Residual attention ResNet50 (stage 2-4) with HRC (stage 3, 4) | 98.39% | 98.54% | | Residual attention ResNet50 (stage 2-4) with HRC (stage 3, 4) | 98.31% | 98.42% | | Residual attention ResNet50 (stage 2-4) with HRC (stage 3, 5) | 98.33% | 98.46% | | Residual attention ResNet50 (stage 2-4) with HRC (stage 3-5) | 98.38% | 98.54% | | Residual attention ResNet50 (stage 2-4) with HRC (stage 4-5) | 98.35% | 98.47% |
|
|
