Face Verification on IJB-A
0.9893TAR @ FAR=1%SFace
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SFaceTraining Dataset=MS1MV2, Backbone=ResNet-1002022.05 | 0.9893 | 0.9802 | 0.9951 | — | — | — | |
| ArcFaceTraining Dataset=MS1MV2, Backbone=ResNet-1002022.05 | 0.9875 | 0.976 | 0.9953 | — | — | — | |
| SFaceTraining Dataset=VGGFace2, Backbone=ResNet-502022.05 | 0.9874 | 0.9685 | 0.9967 | — | — | — | |
| ArcFaceTraining Dataset=VGGFace2, Backbone=ResNet-502022.05 | 0.9864 | 0.9624 | 0.9951 | — | — | — | |
| L2-Face2022.05 | 0.97 | 0.943 | 0.984 | — | — | — | |
| UniformFace2022.05 | 0.969 | 0.923 | — | — | — | — | |
| Crystal Loss2022.05 | 0.969 | 0.949 | 0.984 | — | — | — | |
| VGGFace22022.05 | 0.968 | 0.921 | 0.99 | — | — | — | |
| FTL2022.05 | 0.953 | 0.912 | — | — | — | — | |
| ResNet-18 (end2end+retrain)Training Data=full MS-Celeb-1M, Strategy=DREAM block, fine-tuned=false2018.03 | 0.944 | 0.868 | — | — | — | — | |
| Quality Aware Network (f.t.)fine-tuned=true2018.03 | 0.942 | 0.893 | — | — | — | — | |
| NAN2022.05 | 0.941 | 0.881 | 0.978 | — | — | — | |
| Crosswhite et al.2016.11 | 0.939 | — | — | — | — | — | |
| Template Adaptation (f.t.)fine-tuned=true2018.03 | 0.939 | — | — | — | — | — | |
| Template Adaption2022.05 | 0.939 | 0.836 | 0.979 | — | — | — | |
| ResNet-18 (naïve)Training Data=full MS-Celeb-1M, Strategy=naïve, fine-tuned=false2018.03 | 0.934 | 0.836 | — | — | — | — | |
| All-In-One CNNTPE=true2016.11 | 0.922 | 0.823 | 0.976 | — | — | — | |
| Sankaranarayanan et al.TPE=true2016.11 | 0.9 | 0.813 | 0.964 | — | — | — | |
| CNNmedia+TPE (f.t.)fine-tuned=true2018.03 | 0.9 | 0.813 | — | — | — | — | |
| NAN2016.11 | 0.897 | 0.785 | 0.959 | — | — | — | |
| All-In-One CNNTPE=false2016.11 | 0.893 | 0.787 | 0.968 | — | — | — | |
| ResNet-50 (end2end+retrain)Training Data=MS-Celeb-1M subset, Strategy=DREAM block, fine-tuned=false2018.03 | 0.891 | 0.764 | — | — | — | — | |
| Masi et al.2016.11 | 0.886 | 0.725 | — | — | — | — | |
| Augmentation+Video Pooling+Rendered Testfine-tuned=false2018.03 | 0.886 | 0.725 | — | — | — | — | |
| ResNet-50 (naïve)Training Data=MS-Celeb-1M subset, Strategy=naïve, fine-tuned=false2018.03 | 0.881 | 0.714 | — | — | — | — | |
| ResNet-18 (end2end+retrain)Training Data=MS-Celeb-1M subset, Strategy=DREAM block, fine-tuned=false2018.03 | 0.872 | 0.712 | — | — | — | — | |
| Sankaranarayanan et al.TPE=false2016.11 | 0.871 | 0.766 | 0.952 | — | — | — | |
| ResNet-18 (naïve)Training Data=MS-Celeb-1M subset, Strategy=naïve, fine-tuned=false2018.03 | 0.84 | 0.656 | — | — | — | — | |
| Chen et al.2016.11 | 0.838 | — | 0.967 | — | — | — | |
| DCNNfusion (f.t.)fine-tuned=true2018.03 | 0.838 | — | — | — | — | — | |
| VGGFace2022.05 | 0.834 | 0.62 | 0.954 | — | — | — | |
| PAMsfine-tuned=false2018.03 | 0.826 | 0.652 | — | — | — | — | |
| Pooling Facesfine-tuned=false2018.03 | 0.819 | 0.631 | — | — | — | — | |
| VGG-Face2016.11 | 0.805 | 0.604 | 0.937 | — | — | — | |
| Deep Multi-Posefine-tuned=false2018.03 | 0.787 | — | — | — | — | — | |
| Multi-task CNNfine-tuned=false2018.03 | 0.787 | — | — | — | — | — | |
| Wang et al.fine-tuned=false2018.03 | 0.729 | 0.51 | — | — | — | — | |
| GOTS2016.11 | 0.41 | 0.208 | 0.63 | — | — | — | |
| CNNpooling=average2016.04 | — | — | — | 0.287 | 0.146 | 0.051 | |
| CNNpooling=media2016.04 | — | — | — | 0.234 | 0.129 | 0.048 | |
| CNN + TPEpooling=media2016.04 | — | — | — | 0.187 | 0.1 | 0.036 | |
| Crosswhite et al.2016.04 | — | — | — | 0.135 | 0.06 | 0.017 | |
| GOTS2016.04 | — | — | — | 0.8 | 0.59 | 0.37 | |
| Masi et al.2016.04 | — | — | — | 0.275 | 0.114 | — | |
| NAN2016.04 | — | — | — | 0.215 | 0.103 | 0.041 | |
| VGG-Face2016.04 | — | — | — | 0.396 | 0.195 | 0.063 |