Face Retrieval on VGGFace2
89.1mAPOriginal
Evaluation results
| Method | Links | ||
|---|---|---|---|
| OriginalBackbone=IResNet-502025.12 | 89.1 | 98.8 | |
| Contrastive Unlearninglambda_retain=02025.12 | 89 | 98.8 | |
| Random Labelinglambda_retain=12025.12 | 88.4 | 98.8 | |
| Gradient Ascentlearning rate=1e-52025.12 | 88.03 | 98.8 | |
| Dispersion Losslearning rate=1e-4, margin=0.2, batch size=1602025.12 | 87.5 | 98.8 | |
| Hard Dispersion Losslearning rate=1e-4, margin=0.2, batch size=1602025.12 | 87.47 | 98.8 | |
| Boundary Shrinklearning rate=1e-52025.12 | 87.03 | 98.73 | |
| Lipschitz Unlearninglambda_retain=0.052025.12 | 86.17 | 98.73 |