Face Identification on IJB-C
96.31Rank-1 AccuracyArcFace+ours (CFSM)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ArcFace+ours (CFSM)Train Data=cleaned MS1MV2, #labeled=3.9M, #unlabeled=70K, Backbone=ResNet-502022.07 | 96.31 | 97.48 | |
| ArcFace (Baseline)Train Data=cleaned MS1MV2, #labeled=3.9M, Backbone=ResNet-502022.07 | 95.89 | 97.08 | |
| AFRNTrain Data=VGGFace2-*, #labeled=3.1M, Backbone=ResNet-1012022.07 | 95.7 | 97.6 | |
| DULTrain Data=MS1M-*, #labeled=3.6M, Backbone=ResNet-642022.07 | 95.7 | 97.6 | |
| Shi et al.Train Data=cleaned MS1MV2, #labeled=3.9M, #unlabeled=70K, Backbone=ResNet-502022.07 | 95.61 | 97.13 | |
| PFETrain Data=MS1M-*, #labeled=4.4M, Backbone=ResNet-642022.07 | 95.49 | 97.17 | |
| VGGFace2Train Data=VGGFace2, #labeled=3.3M, Backbone=SE-ResNet-502022.07 | 95.1 | — | |
| ArcFaceTrain Data=MS1MV2, #labeled=5.8M, Backbone=ResNet-502022.07 | 93.26 | 95.33 |