Face Verification on YouTube Face (YTF) 40 (10-fold cross-validation)
97.3AccuracyVGG Face
Evaluation results
| Method | Links | ||
|---|---|---|---|
| VGG FaceAd.Tr.=Y, Protocol=unrestricted2017.03 | 97.3 | — | |
| VGG FaceImages=2.6M, #nets=1, One loss=No2017.03 | 97.3 | — | |
| DeepVisageAd.Tr.=N, Protocol=restricted2017.03 | 96.24 | — | |
| DeepVisageImages=4.48M, #nets=1, One loss=Yes2017.03 | 96.24 | — | |
| L2-S (RX101)Images=3.7M, #nets=1, One loss=Yes, Backbone=ResNeXt-101, Loss=L2-softmax2017.03 | 96.08 | — | |
| L2-S (R101)Images=3.7M, #nets=1, One loss=Yes, Backbone=ResNet-101, Loss=L2-softmax2017.03 | 96.02 | — | |
| NANAd.Tr.=Y, Protocol=unrestricted2017.03 | 95.72 | — | |
| NANImages=3M, #nets=1, One loss=No2017.03 | 95.72 | — | |
| L2-S (FR)Images=3.7M, #nets=1, One loss=Yes, Backbone=Face-Resnet, Loss=L2-softmax2017.03 | 95.54 | — | |
| FaceNetAd.Tr.=N, Protocol=restricted2017.03 | 95.18 | — | |
| FaceNetImages=200M, #nets=1, One loss=Yes2017.03 | 95.12 | — | |
| SphereFaceImages=0.5M, #nets=1, One loss=Yes2017.03 | 95 | — | |
| Center LossAd.Tr.=N, Protocol=restricted2017.03 | 94.9 | — | |
| Wen et al.Images=0.7M, #nets=1, One loss=No2017.03 | 94.9 | — | |
| softmax(FR)Images=3.7M, #nets=1, One loss=Yes, Backbone=Face-Resnet2017.03 | 93.82 | — | |
| Sparse ConvNetAd.Tr.=N, Protocol=restricted2017.03 | 93.5 | — | |
| MFM-CNNAd.Tr.=N, Protocol=restricted2017.03 | 93.4 | — | |
| DeepID2+Ad.Tr.=N, Protocol=restricted2017.03 | 93.2 | — | |
| DeepID-2+Images=-, #nets=25, One loss=No2017.03 | 93.2 | — | |
| CASIA-WebfaceAd.Tr.=Y, Protocol=restricted2017.03 | 92.24 | — | |
| VGG FaceAd.Tr.=N, Protocol=restricted2017.03 | 91.6 | — | |
| DeepfaceAd.Tr.=Y, Protocol=restricted2017.03 | 91.4 | — | |
| Deep FaceImages=4M, #nets=3, One loss=No2017.03 | 91.4 | — | |
| DeepFace-singleprotocol=1:1, validation=10-fold cross-validation2017.04 | — | 96.3 | |
| EigenPEPprotocol=1:1, validation=10-fold cross-validation2017.04 | — | 92.6 | |
| NANprotocol=1:1, validation=10-fold cross-validation2017.04 | — | 98.7 |