Fair Classification on CelebA (clean)
86.32Accuracy (s+)MSAM
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| MSAMBackbone=ResNet-18, Target attribute=Big Nose, Sensitive attribute=Age, Test Data=clean2025.03 | 86.32 | 0.0128 | 64.22 | 0.0756 | 82.79 | |
| SAMBackbone=ResNet-18, Target attribute=Big Nose, Sensitive attribute=Age, Test Data=clean2025.03 | 85.9 | 0.009 | 70.43 | 0.0666 | 82.15 | |
| FairSAMBackbone=ResNet-18, Target attribute=Big Nose, Sensitive attribute=Age, Test Data=clean2025.03 | 85.74 | 0.0399 | 74.8 | 0.0499 | 83.1 | |
| VanillaBackbone=ResNet-18, Target attribute=Big Nose, Sensitive attribute=Age, Test Data=clean2025.03 | 85.72 | 0.0115 | 71.71 | 0.0862 | 82.32 | |
| GroupSAMBackbone=ResNet-18, Target attribute=Big Nose, Sensitive attribute=Age, Test Data=clean2025.03 | 85.71 | 0.0074 | 70.46 | 0.0534 | 81.99 | |
| ReweighedBackbone=ResNet-18, Target attribute=Big Nose, Sensitive attribute=Age, Test Data=clean2025.03 | 85.27 | 0.0436 | 71.56 | 0.0836 | 79.83 | |
| GroupDROBackbone=ResNet-18, Target attribute=Big Nose, Sensitive attribute=Age, Test Data=clean2025.03 | 79.79 | 0.0071 | 71.44 | 0.0509 | 77.22 | |
| FairRegBackbone=ResNet-18, Target attribute=Big Nose, Sensitive attribute=Age, Test Data=clean2025.03 | 65.3 | 0.0215 | 64.92 | 0.0588 | 65.17 |