Face Retrieval on CFP-FP
91mAPOriginal
Evaluation results
| Method | Links | ||
|---|---|---|---|
| OriginalBackbone=IResNet-502025.12 | 91 | 97.5 | |
| Contrastive Unlearninglambda_retain=02025.12 | 91 | 97.57 | |
| Random Labelinglambda_retain=12025.12 | 90 | 97.33 | |
| Gradient Ascentlearning rate=1e-52025.12 | 89.4 | 97.2 | |
| Dispersion Losslearning rate=1e-4, margin=0.2, batch size=322025.12 | 89.2 | 97.3 | |
| Hard Dispersion Losslearning rate=1e-4, margin=0.2, batch size=322025.12 | 89.2 | 97.27 | |
| Boundary Shrinklearning rate=1e-52025.12 | 88.47 | 97.13 | |
| Lipschitz Unlearninglambda_retain=0.052025.12 | 87.67 | 97.07 | |
| Originalbatch size=1602025.12 | 70.1 | 74.1 | |
| Dispersion Lossbatch size=1602025.12 | 63.23 | 72.17 | |
| Hard Dispersion Lossbatch size=1602025.12 | 57.3 | 69.7 |