AI-generated image detection on CycleGAN generator (test)
99.99APOurs (ViT-L/14)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Ours (ViT-L/14)Training Dataset=ProGAN, Backbone=ViT-L/142026.01 | 99.99 | 99.62 | |
| FatFormerTraining Dataset=ProGAN2026.01 | 99.85 | 95.87 | |
| SAFETraining Dataset=ProGAN2026.01 | 99.71 | 98.22 | |
| Ours (ViT-B/32)Training Dataset=ProGAN, Backbone=ViT-B/322026.01 | 99.61 | 97.05 | |
| UnivFDTraining Dataset=ProGAN2026.01 | 99.46 | 98.33 | |
| VIB-NetTraining Data=Diffusion-based data (SDv1.4 and VQDM)2026.04 | 98.95 | — | |
| EffortTraining Data=Diffusion-based data (SDv1.4 and VQDM)2026.04 | 97.98 | — | |
| UnivfdTraining Data=Diffusion-based data (SDv1.4 and VQDM)2026.04 | 97.69 | — | |
| NPRTraining Dataset=ProGAN2026.01 | 97.63 | 90.27 | |
| CLIPpingTraining Data=Diffusion-based data (SDv1.4 and VQDM)2026.04 | 96.66 | — | |
| MAFLTraining Data=Diffusion-based data (SDv1.4 and VQDM)2026.04 | 95.87 | — | |
| LNPTraining Dataset=ProGAN2026.01 | 94.06 | 83.54 | |
| LGradTraining Dataset=ProGAN2026.01 | 93.84 | 85.96 | |
| CNNSpotTraining Dataset=ProGAN2026.01 | 90.07 | 82.85 | |
| FreDectTraining Dataset=ProGAN2026.01 | 84.78 | 78.77 | |
| NPRTraining Data=Diffusion-based data (SDv1.4 and VQDM)2026.04 | 81.1 | — | |
| GramNetTraining Dataset=ProGAN2026.01 | 74.82 | 73.73 |