AI-generated image detection on AIGIBench R3GAN
95.4AccuracyC2P-CLIP
Evaluation results
| Method | Links | ||
|---|---|---|---|
| C2P-CLIPevaluation_strategy=cross-model2025.01 | 95.4 | 99.8 | |
| DFFreqevaluation_strategy=cross-model2025.01 | 94.6 | 94.9 | |
| FatFormerevaluation_strategy=cross-model2025.01 | 94.4 | 99.3 | |
| UnivFDevaluation_strategy=cross-model2025.01 | 93 | 98.6 | |
| AIDEVenue=ICLR’252026.05 | 92.9 | 97.1 | |
| LEGOVenue=-2026.05 | 90.8 | 96.7 | |
| EffortVenue=ICML’252026.05 | 90.2 | 95.8 | |
| AIDEevaluation_strategy=cross-model2025.01 | 89 | 94.8 | |
| CO-SPYVenue=CVPR’252026.05 | 84.5 | 95.9 | |
| CLIPDetectionVenue=CVPR’232026.05 | 83.5 | 91.2 | |
| NPRevaluation_strategy=cross-model2025.01 | 80.8 | 75.6 | |
| LaDeDaevaluation_strategy=cross-model2025.01 | 72.9 | 69.2 | |
| LGradevaluation_strategy=cross-model2025.01 | 65.6 | 69.1 | |
| FreqNetevaluation_strategy=cross-model2025.01 | 62.3 | 56.8 | |
| FreqNetVenue=AAAI’232026.05 | 62.3 | 56.8 | |
| DFFreqVenue=arXiv’252026.05 | 61.7 | 74 | |
| LaDeDaVenue=arXiv’242026.05 | 54.8 | 72.6 | |
| LGradVenue=CVPR’232026.05 | 54.4 | 58.7 | |
| NPRVenue=CVPR’242026.05 | 50.8 | 61.1 | |
| CNNDetectionVenue=CVPR’202026.05 | 50.4 | 52.7 | |
| CNN-Spotevaluation_strategy=cross-model2025.01 | 49.9 | 48.4 | |
| ResNet-50Venue=CVPR’162026.05 | 48.5 | 53.7 | |
| Gram-netVenue=CVPR’202026.05 | 47.9 | 52.5 |