Table 4.
The accuracy rates of SVMs trained on embeddings extracted from students based on standard teacher–student (TS) or triplet loss (TL) strategies
| Network | Method | FER+ (%) | AffectNet (%) | UTKFace (%) |
|---|---|---|---|---|
| VGG-f | TS | 80.17 | 48.75 | 89.13 |
| TL | 80.05 | 48.13 | 89.55 | |
| TS+SVM | 80.39 | 48.52 | 89.04 | |
| TL+SVM | 79.06 | 47.01 | 89.70 | |
| TS+TL+SVM | 81.09 | 48.70 | 89.82 | |
| VGG-face | TS | 82.37 | 49.75 | 88.45 |
| TL | 82.57 | 49.71 | 88.31 | |
| TS+SVM | 82.34 | 48.89 | 90.35 | |
| TL+SVM | 82.37 | 49.90 | 90.27 | |
| TS+TL+SVM | 82.75 | 50.09 | 90.35 |
These models are compared with SVMs trained on concatenated embeddings as well as the students providing the embeddings. Results are reported for two tasks: facial expression recognition (on FER+ and AffectNet) and gender prediction (on UTKFace)