AI-generated Image Detection on FakeForm BLIP-3o
91.5AccuracyAIGI-Holmes
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| AIGI-HolmesBackbone=ViT, FPS=2.2, Params (M)=> 70002026.05 | 91.5 | 93 | |
| AIDEBackbone=ResNeXt, FPS=18.9, FLOPs=225.69, Params (M)=893.542026.05 | 91.2 | 93.3 | |
| FreqNetBackbone=ResNet, FPS=157.2, FLOPs=1.97, Params (M)=1.852026.05 | 89.5 | 91.7 | |
| CoDABackbone=ResNet, FPS=125.2, FLOPs=2.37, Params (M)=1.482026.05 | 88.3 | 91.8 | |
| SAFEBackbone=ResNet, FPS=180.4, FLOPs=2.29, Params (M)=1.442026.05 | 84.4 | 92.2 | |
| UFD2026.05 | 76.6 | 86 | |
| Humans2026.05 | 74.2 | 82.4 | |
| FatFormerBackbone=ViT, FPS=13.3, FLOPs=216.83, Params (M)=492.592026.05 | 67.1 | 92.6 | |
| CNNDetBackbone=Xception, FPS=26.2, FLOPs=16.57, Params (M)=23.512026.05 | 63.5 | 78.4 | |
| NPRBackbone=ResNet, FPS=200.6, FLOPs=1.75, Params (M)=1.442026.05 | 62.1 | 67.7 |