AI-generated image detection on Stable Diffusion (val)
99.79AccuracyResNet50
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| ResNet50Attack method=No attack, Backbone=ResNet502025.12 | 99.79 | 99.76 | 99.82 | 99.79 | |
| ResNet50Attack method=FGSM, Epsilon (ε)=0.05, Backbone=ResNet502025.12 | 50.88 | 50.46 | 95.25 | 65.97 | |
| ResNet50Attack method=FGSM, Epsilon (ε)=0.03, Backbone=ResNet502025.12 | 48.2 | 49.02 | 89.86 | 63.43 | |
| ResNet50Attack method=FGSM, Epsilon (ε)=0.01, Backbone=ResNet502025.12 | 44.71 | 46.99 | 82.77 | 59.95 | |
| ResNet50Attack method=PGD, Epsilon (ε)=0.05, Backbone=ResNet502025.12 | 40.57 | 44.8 | 81.15 | 57.73 | |
| ResNet50Attack method=PGD, Epsilon (ε)=0.03, Backbone=ResNet502025.12 | 40.28 | 44.61 | 80.47 | 57.4 | |
| ResNet50Attack method=PGD, Epsilon (ε)=0.01, Backbone=ResNet502025.12 | 35.45 | 41.41 | 70.11 | 52.07 |