Face Verification Suite (LFW, AgeDB, CALFW, CPLFW, CFP-FP) (10-fold CV)
97.19Average AccuracyAdaFace
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| AdaFace# Synthetic images=0, # Real images=5.8M, Backbone=ResNet100, Training Data=MS1MV22022.10 | 97.19 | 99.82 | 98.05 | 96.08 | 93.53 | 98.49 | — | — | 97.28 | — | |
| MagFace# Synthetic images=0, # Real images=5.8M, Backbone=ResNet100, Training Data=MS1MV22022.10 | 97.1 | 99.83 | 98.17 | 96.15 | 92.87 | 98.46 | — | — | 97.05 | — | |
| ArcFace# Synthetic images=0, # Real images=5.8M, Backbone=ResNet100, Training Data=MS1MV22022.10 | 96.99 | 99.81 | 98.05 | 95.96 | 92.72 | 98.4 | — | — | 96.98 | — | |
| CosFace# Synthetic images=0, # Real images=5.8M, Backbone=ResNet100, Training Data=MS1MV22022.10 | 96.91 | 99.78 | 98.17 | 96.18 | 92.18 | 98.26 | — | — | 96.74 | — | |
| SphereFace# Synthetic images=0, # Real images=5.8M, Backbone=ResNet100, Training Data=MS1MV22022.10 | 96.08 | 99.67 | 97.05 | 95.58 | 91.27 | 96.84 | — | — | 95.93 | — | |
| SFaceBackbone=ResNet50, Training Dataset=CASIA-WebFace, a=0.87, b=1.22022.05 | 94.93 | — | — | — | — | — | — | — | — | — | |
| CosFaceBackbone=ResNet50, Training Dataset=CASIA-WebFace, m=0.352022.05 | 94.91 | — | — | — | — | — | — | — | — | — | |
| ArcFaceBackbone=ResNet50, Training Dataset=CASIA-WebFace, m=0.42022.05 | 94.91 | — | — | — | — | — | — | — | — | — | |
| CombinedBackbone=ResNet50, Training Dataset=CASIA-WebFace, m=0.9,0.4,0.152022.05 | 94.9 | — | — | — | — | — | — | — | — | — | |
| SFaceBackbone=ResNet50, Training Dataset=CASIA-WebFace, a=0.9, b=1.22022.05 | 94.88 | — | — | — | — | — | — | — | — | — | |
| SFaceBackbone=ResNet50, Training Dataset=CASIA-WebFace, a=0.93, b=1.22022.05 | 94.88 | — | — | — | — | — | — | — | — | — | |
| ArcFaceBackbone=ResNet50, Training Dataset=CASIA-WebFace, m=0.52022.05 | 94.83 | — | — | — | — | — | — | — | — | — | |
| SV-AM-Softmax# Synthetic images=0, # Real images=5.8M, Backbone=ResNet100, Training Data=MS1MV22022.10 | 94.83 | 99.5 | 95.68 | 94.38 | 89.48 | 95.1 | — | — | 94.69 | — | |
| SFaceBackbone=ResNet50, Training Dataset=CASIA-WebFace, a=0.9, b=1.32022.05 | 94.8 | — | — | — | — | — | — | — | — | — | |
| ArcFaceBackbone=ResNet50, Training Dataset=CASIA-WebFace, m=0.32022.05 | 94.65 | — | — | — | — | — | — | — | — | — | |
| CASIA-WebFace (Real)Training Data (# images, # IDs x # imgs/ID)=0.49M (approx. 10.5K x 47), Backbone=IR-SE50 + AdaFace2023.04 | 94.62 | 99.42 | 94.08 | 93.32 | 89.73 | 96.56 | — | — | — | 0 | |
| D-softmaxBackbone=ResNet50, Training Dataset=CASIA-WebFace, d=0.92022.05 | 94.29 | — | — | — | — | — | — | — | — | — | |
| softmaxBackbone=ResNet50, Training Dataset=CASIA-WebFace2022.05 | 93.82 | — | — | — | — | — | — | — | — | — | |
| NSoftmaxBackbone=ResNet50, Training Dataset=CASIA-WebFace, s=202022.05 | 93.72 | — | — | — | — | — | — | — | — | — | |
| DigiFace-1M (SX+Real best)# Synthetic images=1.22M, # Real images=120K, Backbone=ResNet1002022.10 | 93.61 | 99.33 | 91.55 | 91.78 | 89.47 | 95.93 | — | — | 94.91 | — | |
| SphereFaceBackbone=ResNet50, Training Dataset=CASIA-WebFace, m=1.352022.05 | 92.99 | — | — | — | — | — | — | — | — | — | |
| CryptoFaceNet16Params=3.78M, #Boot=1, Resolution=128x1282025.08 | 91.46 | 98.78 | 92.9 | 93.73 | 83.95 | 87.94 | 1,446 | — | — | — | |
| DCFaceTraining Data (# images, # IDs x # imgs/ID)=1.2M (20K x 50 + 40K x 5), Backbone=IR-SE50 + AdaFace2023.04 | 91.21 | 98.58 | 90.97 | 92.82 | 85.07 | 88.61 | — | — | — | 3.74 | |
| CryptoFaceNet9Params=2.12M, #Boot=1, Resolution=96x962025.08 | 90.99 | 99.18 | 91.38 | 93.32 | 84.23 | 86.81 | 1,395 | — | — | — | |
| DCFaceTraining Data (# images, # IDs x # imgs/ID)=1.0M (20K x 50), Backbone=IR-SE50 + AdaFace2023.04 | 90.86 | 98.83 | 90.45 | 92.38 | 84.22 | 88.4 | — | — | — | 4.14 | |
| MPCNNBackbone=ResNet44, Params=0.72M, #Boot=43, Resolution=64x642025.08 | 89.64 | 98.27 | 87.45 | 90.85 | 83.72 | 87.9 | 9,845 | — | — | — | |
| DCFaceTraining Data (# images, # IDs x # imgs/ID)=0.5M (10K x 50), Backbone=IR-SE50 + AdaFace2023.04 | 89.56 | 98.55 | 89.7 | 91.6 | 82.62 | 85.33 | — | — | — | 5.65 | |
| CryptoFaceNet4Params=0.94M, #Boot=1, Resolution=64x642025.08 | 89.42 | 98.87 | 89.45 | 91.6 | 81.98 | 85.21 | 1,364 | — | — | — | |
| DigiFace-1M (SX best)# Synthetic images=1.22M, # Real images=0, Backbone=ResNet1002022.10 | 86.37 | 96.17 | 81.1 | 82.55 | 82.23 | 89.81 | — | — | 89.4 | — | |
| DigiFaceVenue=WACV23, Training Data (# images, # IDs x # imgs/ID)=1.2M (10K x 72 + 100K x 5), Backbone=IR-SE50 + AdaFace2023.04 | 86.37 | 96.17 | 81.1 | 82.55 | 82.23 | 89.81 | — | — | — | 9.55 | |
| MPCNNBackbone=ResNet32, Params=0.53M, #Boot=31, Resolution=64x642025.08 | 85.6 | 97.02 | 83.02 | 87 | 78.9 | 82.07 | 7,367 | — | — | — | |
| DigiFaceVenue=WACV23, Training Data (# images, # IDs x # imgs/ID)=0.5M (10K x 50), Backbone=IR-SE50 + AdaFace2023.04 | 83.45 | 95.4 | 76.97 | 78.62 | 78.87 | 87.4 | — | — | — | 13.39 | |
| AutoFHEBackbone=ResNet32, Params=0.53M, #Boot=8, Resolution=64x642025.08 | 82.69 | 93.53 | 80.88 | 85.4 | 75.67 | 77.96 | 4,001 | — | — | — | |
| SynFaceVenue=ICCV21, Training Data (# images, # IDs x # imgs/ID)=0.5M (10K x 50), Backbone=IR-SE50 + AdaFace2023.04 | 74.75 | 91.93 | 61.63 | 74.73 | 70.43 | 75.03 | — | — | — | 26.58 |