Category-Agnostic Perceptual Artifact Detection on AIGI Detection Dataset Category-Agnostic PAD
2,730IoUMulti-task learning
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Multi-task learningBackbone=Swin Transformer2026.01 | 2,730 | 36.2 | 52.5 | 42.8 | |
| PAD-onlyBackbone=Swin Transformer2026.01 | 2,720 | 40.4 | 45.4 | 42.7 | |
| DRCT-ConvBEvaluation Protocol=Binarized interpretation heatmaps2026.01 | 900 | 14.8 | 18.8 | 16.6 | |
| Gram-NetEvaluation Protocol=Binarized interpretation heatmaps2026.01 | 580 | 16.9 | 8.1 | 11 | |
| CoDEEvaluation Protocol=Binarized interpretation heatmaps2026.01 | 290 | 11.5 | 3.8 | 5.7 | |
| CNNSpotEvaluation Protocol=Binarized interpretation heatmaps2026.01 | 90 | 7.7 | 1 | 1.8 | |
| DRCT-CLIPEvaluation Protocol=Binarized interpretation heatmaps2026.01 | 70 | 13.7 | 0.8 | 1.5 | |
| FatFormerEvaluation Protocol=Binarized interpretation heatmaps2026.01 | 50 | 10.2 | 0.5 | 0.9 | |
| UnivFDEvaluation Protocol=Binarized interpretation heatmaps2026.01 | 40 | 10.1 | 0.4 | 0.8 |