Authenticity Judgment on AIGI Detection Dataset Authenticity Judgment
89.9AccuracyTransfer learning
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Transfer learningBackbone=Swin Transformer, Evaluation Protocol=full fine-tuning2026.01 | 89.9 | 93.5 | 91.6 | 92.5 | |
| AJ-onlyBackbone=Swin Transformer2026.01 | 89.3 | 96.3 | 85.8 | 90.2 | |
| Multi-task learningBackbone=Swin Transformer2026.01 | 89.1 | 92.6 | 92 | 92.3 | |
| DRCT-ConvB2026.01 | 82.5 | 88.4 | 74.8 | 81 | |
| Transfer learningBackbone=Swin Transformer, Evaluation Protocol=linear probing2026.01 | 82.1 | 93.5 | 73.9 | 82.5 | |
| CoDE2026.01 | 76.5 | 93.1 | 57.2 | 70.9 | |
| Gram-Net2026.01 | 62.7 | 57.7 | 95 | 71.8 | |
| DRCT-CLIP2026.01 | 53 | 52.2 | 71.4 | 60.3 | |
| FatFormer2026.01 | 52.1 | 57.3 | 16.3 | 25.4 | |
| UnivFD2026.01 | 50.2 | 64.1 | 1 | 2 | |
| CNNSpot2026.01 | 48.6 | 39.7 | 5.6 | 9.8 |