Face Verification on LFW (AUROC, TPR, and PE Gap Metrics)
99.42AUROCAC-Ridge
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| AC-RidgeBackbone=FaceNet, Demographic label requirement at inference=True2026.06 | 99.42 | 76.47 | 92.63 | 1.25 | 4.32 | 3.28 | 2.48 | 97.21 | 1 | |
| AC-LinearBackbone=FaceNet, Demographic label requirement at inference=True2026.06 | 99.39 | 80.77 | 93.1 | 3.26 | 4.77 | 2.85 | 2.33 | 97.06 | 7 | |
| AC-RidgeBackbone=MagFace, Demographic labels at inference=false2026.06 | 99.37 | 75.9 | 94.03 | 1.9 | 3.66 | 4.98 | 2.77 | 98.18 | 2 | |
| AC-RidgeBackbone=AdaFace, Demographic labels at inference=false2026.06 | 99.37 | 75.9 | 94.03 | 1.9 | 3.66 | 4.98 | 2.77 | 98.18 | 2 | |
| AC-LinearBackbone=MagFace, Demographic labels at inference=false2026.06 | 99.34 | 83.33 | 93.77 | 1.9 | 3.22 | 4.02 | 2.62 | 98.42 | 3 | |
| AC-LinearBackbone=AdaFace, Demographic labels at inference=false2026.06 | 99.34 | 83.33 | 93.77 | 1.9 | 3.22 | 4.02 | 2.62 | 98.42 | 3 | |
| FairCalBackbone=MagFace, Demographic labels at inference=false2026.06 | 99.32 | 83.47 | 92.67 | 2.33 | 4.78 | 2.04 | 2.56 | 97.59 | 1 | |
| FairCalBackbone=AdaFace, Demographic labels at inference=false2026.06 | 99.32 | 83.47 | 92.67 | 2.33 | 4.78 | 2.04 | 2.56 | 97.59 | 1 | |
| FALCONBackbone=FaceNet, Demographic label requirement at inference=False2026.06 | 99.29 | 82.8 | 92.4 | 1.25 | 4.45 | 25.55 | 14.75 | 94.68 | 7 | |
| DemoNormBackbone=FaceNet, Demographic label requirement at inference=True2026.06 | 99.28 | 82.3 | 92.53 | 1.59 | 4.57 | 15.66 | 13.2 | 94.65 | — | |
| BaselineBackbone=FaceNet, Demographic label requirement at inference=False2026.06 | 99.27 | 81.4 | 91.87 | 1.25 | 4.32 | 17.1 | 7.43 | 94.65 | — | |
| OracleBackbone=FaceNet, Demographic label requirement at inference=True2026.06 | 99.22 | 81.23 | 91.13 | 1.92 | 5.45 | 2.19 | 2.53 | 93.51 | — | |
| GSTBackbone=FaceNet, Demographic label requirement at inference=True2026.06 | 99.21 | 81.4 | 92.27 | 1.49 | 4.09 | 2.21 | 2.62 | 94.65 | — | |
| FairCalBackbone=FaceNet, Demographic label requirement at inference=False2026.06 | 99.1 | 65.4 | 91 | 1.89 | 3.47 | 2.04 | 2.58 | 96.56 | 0 | |
| BaselineBackbone=MagFace, Demographic labels at inference=false2026.06 | 98.96 | 87 | 93.27 | 2.14 | 5.08 | 25.72 | 15.82 | 97.25 | — | |
| BaselineBackbone=AdaFace, Demographic labels at inference=false2026.06 | 98.96 | 87 | 93.27 | 2.14 | 5.08 | 25.72 | 15.82 | 97.25 | — | |
| Oracle†Backbone=AdaFace, Demographic labels at inference=true2026.06 | 98.96 | 85.83 | 93.13 | 2.94 | 5.13 | 2.17 | 2.82 | 97.41 | — | |
| OracleBackbone=ArcFace R100, Demographic labels at inference=true2026.06 | 98.95 | 72.27 | 87.43 | 0.16 | 1.37 | 0.6 | 2.91 | 97.88 | — | |
| GST†Backbone=AdaFace, Demographic labels at inference=true2026.06 | 98.95 | 86.27 | 92.9 | 2.14 | 5.08 | 2.13 | 2.82 | 97.25 | — | |
| DemoNorm†Backbone=AdaFace, Demographic labels at inference=true2026.06 | 98.95 | 87.1 | 93.27 | 2.14 | 6.28 | 14.86 | 12.79 | 97.25 | — | |
| OracleBackbone=ArcFace R100, Demographic labels required at inference=True2026.06 | 98.95 | 72.27 | 87.43 | 0.16 | 1.37 | 0.6 | 2.91 | 97.88 | — | |
| FALCONBackbone=AdaFace, Demographic labels at inference=false2026.06 | 98.92 | 87.3 | 93.2 | 2.14 | 5.08 | 30.03 | 16.88 | 97.12 | 1 | |
| AC-LinearBackbone=ArcFace R100, Demographic labels at inference=false2026.06 | 98.91 | 72.47 | 87.65 | 0.26 | 1.87 | 0.82 | 2.9 | 97.75 | 1 | |
| AC-RidgeBackbone=ArcFace R50, Demographic-aware=true2026.06 | 98.91 | 96.3 | 96.43 | 1.47 | 5.1 | 2.78 | 1.98 | 92.91 | 0 | |
| AC-LinearBackbone=ArcFace R100, Demographic labels required at inference=False2026.06 | 98.91 | 72.47 | 87.65 | 0.26 | 1.87 | 0.82 | 2.9 | 97.75 | 1 | |
| FairCalBackbone=ArcFace R100, Demographic labels at inference=false2026.06 | 98.9 | 73.15 | 87.78 | 0.21 | 1.57 | 0.39 | 2.87 | 97.77 | 1 | |
| FairCalBackbone=ArcFace R100, Demographic labels required at inference=False2026.06 | 98.9 | 73.15 | 87.78 | 0.21 | 1.57 | 0.39 | 2.87 | 97.77 | 1 | |
| FRAPPEBackbone=ArcFace R100, Demographic labels at inference=false2026.06 | 98.86 | 71.26 | 87.13 | 0.23 | 1.65 | 0.33 | 2.95 | 97.72 | 1 | |
| FRAPPEBackbone=ArcFace R100, Demographic labels required at inference=False2026.06 | 98.86 | 71.26 | 87.13 | 0.23 | 1.65 | 0.33 | 2.95 | 97.72 | 1 | |
| FALCONBackbone=ArcFace R100, Demographic labels at inference=false2026.06 | 98.85 | 70.37 | 86.62 | 0.26 | 1.99 | 14.04 | 8.26 | 97.77 | 1 | |
| FSNBackbone=ArcFace R100, Demographic labels at inference=false2026.06 | 98.85 | 71.82 | 87.12 | 0.17 | 1.47 | 13.05 | 5.69 | 97.68 | 1 | |
| FALCONBackbone=ArcFace R100, Demographic labels required at inference=False2026.06 | 98.85 | 70.37 | 86.62 | 0.26 | 1.99 | 14.04 | 8.26 | 97.77 | 1 | |
| FSNBackbone=ArcFace R100, Demographic labels required at inference=False2026.06 | 98.85 | 71.82 | 87.12 | 0.17 | 1.47 | 13.05 | 5.69 | 97.68 | 1 | |
| FRAPPEBackbone=MagFace, Demographic labels at inference=false2026.06 | 98.84 | 84.97 | 91.5 | 2.12 | 3.01 | 2.1 | 3.18 | 96.96 | 2 | |
| FRAPPEBackbone=AdaFace, Demographic labels at inference=false2026.06 | 98.84 | 84.97 | 91.5 | 2.12 | 3.01 | 2.1 | 3.18 | 96.96 | 2 | |
| BaselineBackbone=ArcFace R100, Demographic labels at inference=false2026.06 | 98.8 | 67.95 | 85.7 | 0.3 | 2.22 | 11.94 | 5.61 | 97.78 | — | |
| DemoNormBackbone=ArcFace R100, Demographic labels at inference=true2026.06 | 98.8 | 70.36 | 86.64 | 0.2 | 1.26 | 22.59 | 19.04 | 97.78 | — | |
| BaselineBackbone=ArcFace R100, Demographic labels required at inference=False2026.06 | 98.8 | 67.95 | 85.7 | 0.3 | 2.22 | 11.94 | 5.61 | 97.78 | — | |
| DemoNormBackbone=ArcFace R100, Demographic labels required at inference=True2026.06 | 98.8 | 70.36 | 86.64 | 0.2 | 1.26 | 22.59 | 19.04 | 97.78 | — | |
| GSTBackbone=ArcFace R100, Demographic labels at inference=true2026.06 | 98.76 | 72 | 87.03 | 0.15 | 1.14 | 0.49 | 3.01 | 97.78 | — | |
| GSTBackbone=ArcFace R100, Demographic labels required at inference=True2026.06 | 98.76 | 72 | 87.03 | 0.15 | 1.14 | 0.49 | 3.01 | 97.78 | — | |
| AC-MLPBackbone=ArcFace R100, Demographic labels at inference=false2026.06 | 98.74 | 70.77 | 86.74 | 0.17 | 1.52 | 1.3 | 3.14 | 97.52 | 1 | |
| AC-MLPBackbone=ArcFace R100, Demographic labels required at inference=False2026.06 | 98.74 | 70.77 | 86.74 | 0.17 | 1.52 | 1.3 | 3.14 | 97.52 | 1 | |
| FRAPPEBackbone=FaceNet, Demographic label requirement at inference=False2026.06 | 98.73 | 63.9 | 83.73 | 1.78 | 4.3 | 2.42 | 3.82 | 93.93 | 0 | |
| AC-RidgeBackbone=ArcFace R100, Demographic labels at inference=false2026.06 | 98.63 | 73.63 | 86.84 | 0.25 | 2.1 | 1.95 | 3.04 | 97.19 | 1 | |
| AC-LinearBackbone=ArcFace R50, Demographic-aware=true2026.06 | 98.63 | 96.23 | 96.5 | 1.43 | 4.24 | 1.64 | 1.87 | 92.15 | 0 | |
| AC-RidgeBackbone=ArcFace R100, Demographic labels required at inference=False2026.06 | 98.63 | 73.63 | 86.84 | 0.25 | 2.1 | 1.95 | 3.04 | 97.19 | 1 | |
| BaselineBackbone=ArcFace R50, Demographic-aware=false2026.06 | 98.53 | 96.3 | 96.47 | 1.43 | 3.66 | 17.07 | 7.57 | 91.77 | — | |
| FALCONBackbone=ArcFace R50, Demographic-aware=false2026.06 | 98.53 | 96.3 | 96.47 | 1.43 | 3.66 | 22.97 | 8.7 | 91.77 | 0 | |
| OracleBackbone=ArcFace R50, Demographic-aware=true2026.06 | 98.53 | 96.3 | 96.47 | 1.43 | 3.66 | 0.85 | 1.83 | 91.77 | — | |
| GSTBackbone=ArcFace R50, Demographic-aware=true2026.06 | 98.53 | 96.3 | 96.47 | 1.43 | 3.66 | 0.85 | 1.83 | 91.77 | — | |
| DemoNormBackbone=ArcFace R50, Demographic-aware=true2026.06 | 98.53 | 96.3 | 96.47 | 1.43 | 3.66 | 15.94 | 13.62 | 91.77 | — | |
| FRAPPEBackbone=ArcFace R50, Demographic-aware=false2026.06 | 98.52 | 96.3 | 96.5 | 1.43 | 3.66 | 0.81 | 1.83 | 91.52 | 1 | |
| FairCalBackbone=ArcFace R50, Demographic-aware=false2026.06 | 98.46 | 96.2 | 96.5 | 1.41 | 4 | 1.23 | 1.9 | 92.18 | 0 | |
| FSNBackbone=ArcFace R50, Demographic-aware=false2026.06 | 98.3 | 96 | 96.23 | 1.57 | 3.96 | 15.49 | 5.19 | 91.04 | 0 | |
| AC-LinearBackbone=AdaFace, Demographic labels at inference=false2026.06 | 98.26 | 61.9 | 82.13 | 1.91 | 3.18 | 4.55 | 4.88 | 93.61 | 8 | |
| AC-MLPBackbone=FaceNet, Demographic label requirement at inference=True2026.06 | 98.03 | 59.83 | 70.53 | 2.17 | 3.86 | 15 | 7.8 | 94.03 | 0 | |
| AC-RidgeBackbone=AdaFace, Demographic labels at inference=false2026.06 | 97.98 | 57.13 | 75.8 | 2.4 | 4.55 | 5.25 | 5.53 | 94.9 | 0 | |
| AC-MLPBackbone=ArcFace R50, Demographic-aware=true2026.06 | 97.96 | 83.1 | 88.67 | 1.37 | 4.44 | 16.88 | 7.23 | 94.94 | 0 | |
| FairCalBackbone=AdaFace, Demographic labels at inference=false2026.06 | 97.86 | 60.4 | 76.67 | 1.34 | 4.68 | 3.43 | 5.39 | 93.36 | 0 | |
| FALCONBackbone=AdaFace, Demographic labels at inference=false2026.06 | 97.29 | 63.77 | 75.93 | 1.25 | 8.83 | 28.04 | 19.23 | 91.71 | 0 | |
| BaselineBackbone=AdaFace, Demographic labels at inference=false2026.06 | 97.12 | 59.77 | 72.3 | 1.65 | 8.98 | 22.57 | 14.94 | 91.6 | — | |
| FSNBackbone=AdaFace, Demographic labels at inference=false2026.06 | 96.7 | 65.7 | 77.37 | 3.24 | 4.06 | 19.06 | 11.68 | 91.9 | 0 | |
| FRAPPEBackbone=AdaFace, Demographic labels at inference=false2026.06 | 96.51 | 57.07 | 73.57 | 2.14 | 4.93 | 3.22 | 6.66 | 90.45 | 0 | |
| FSNBackbone=FaceNet, Demographic label requirement at inference=False2026.06 | 95.58 | 32.37 | 43.43 | 4.19 | 6.24 | 13.28 | 11.03 | 85.36 | 0 | |
| FSNBackbone=AdaFace, Demographic labels at inference=false2026.06 | 95.34 | 48.7 | 63.47 | 3.58 | 5.73 | 18.89 | 12.08 | 90.55 | 0 | |
| AC-MLPBackbone=AdaFace, Demographic labels at inference=false2026.06 | 95.2 | 44.47 | 58.4 | 3.44 | 10.34 | 15.42 | 11.05 | 86.56 | 0 |