Face Verification on IJB-A (test)
0.972TAR @ FAR=0.01SE-GV-3
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| SE-GV-3Training dataset=VF2, D=1282018.10 | 0.972 | — | 0.935 | — | |
| SE-GV-4-g1Training dataset=VF2, D=1282018.10 | 0.972 | — | 0.935 | — | |
| L2-softmaxBackbone=ResNeXt-101, Template Pooling Encoding=true2017.03 | 0.97 | 0.909 | 0.943 | 0.984 | |
| L2-softmaxBackbone=ResNet-101, Template Pooling Encoding=true2017.03 | 0.969 | 0.898 | 0.942 | 0.983 | |
| L2-softmaxBackbone=ResNeXt-1012017.03 | 0.968 | 0.883 | 0.938 | 0.987 | |
| SETraining dataset=MS+VF2, D=20482018.10 | 0.968 | — | 0.921 | — | |
| Chen et al.2017.03 | 0.967 | — | 0.838 | — | |
| L2-softmaxBackbone=ResNet-1012017.03 | 0.967 | 0.879 | 0.937 | 0.984 | |
| MN-vcTraining dataset=VF2, D=20482018.10 | 0.962 | — | 0.92 | — | |
| TDFF2017.03 | 0.961 | 0.875 | 0.919 | 0.988 | |
| TDFF + TPETemplate Pooling Encoding=true2017.03 | 0.961 | 0.877 | 0.921 | 0.989 | |
| SETraining dataset=VF2, D=20482018.10 | 0.958 | — | 0.904 | — | |
| Deep Multi-Pose2017.03 | 0.954 | — | 0.876 | — | |
| L2-softmaxProtocol=FR2017.03 | 0.952 | 0.832 | 0.906 | 0.981 | |
| L2-softmaxProtocol=FR, Template Pooling Encoding=true2017.03 | 0.951 | 0.863 | 0.91 | 0.979 | |
| DREAMTraining dataset=MS2018.10 | 0.944 | — | 0.868 | — | |
| QANTraining dataset=VF+priv22018.10 | 0.942 | — | 0.893 | — | |
| NAN2017.03 | 0.941 | — | 0.881 | 0.979 | |
| NANTraining dataset=priv1, D=1282018.10 | 0.941 | — | 0.881 | — | |
| Template Adaptation2017.03 | 0.939 | — | 0.836 | 0.979 | |
| softmaxProtocol=FR2017.03 | 0.926 | 0.73 | 0.851 | 0.972 | |
| All-In-One Face2017.03 | 0.922 | — | 0.823 | 0.976 | |
| Triplet Embedding2017.03 | 0.9 | — | 0.813 | 0.964 | |
| Masi et al.2017.03 | 0.886 | — | 0.725 | — | |
| AugNet3D Shape=No, Texture=No2016.12 | 0.886 | — | — | — | |
| Pose-Aware Models2017.03 | 0.826 | — | 0.652 | — | |
| VGG-Face2017.03 | 0.805 | — | 0.604 | 0.937 | |
| DCNN_manual+metric2017.03 | 0.787 | — | — | 0.947 | |
| LSFS2017.03 | 0.733 | — | 0.514 | 0.895 | |
| CNN3DMM3D Shape=Yes, Texture=Yes, Protocol=pool2016.12 | 0.6 | — | — | 87 | |
| CNN3DMM3D Shape=Yes, Texture=No, Protocol=pool2016.12 | 0.559 | — | — | 86 | |
| CNN3DMM3D Shape=No, Texture=Yes, Protocol=pool2016.12 | 0.503 | — | — | 83.5 | |
| 3DMM*+pool3D Shape=Yes, Texture=Yes2016.12 | 0.466 | — | — | 75.4 | |
| GOTS2017.03 | 0.41 | — | 0.2 | 0.63 | |
| 3DMM*+pool3D Shape=No, Texture=Yes2016.12 | 0.395 | — | — | 71.1 | |
| 3DMM*+pool3D Shape=Yes, Texture=No2016.12 | 0.306 | — | — | 60.7 | |
| 3DDFA+pool3D Shape=Yes, Texture=No2016.12 | 0.125 | — | — | 43.3 | |
| BinTraining dataset=ImNet+CAS, D=40962018.10 | — | 0.631 | 0.819 | — | |
| SF+RTraining dataset=MS-clean, D=512-pi2018.10 | — | — | 0.932 | — |