Table 2.
Cross-dataset generalization.
| Method | Celeb-DF (AUC) |
|---|---|
| Xception (Rössler et al., 2019) | 66.9 |
| FreqNet (Durall et al., 2020) | 72.7 |
| SRM (Guo et al., 2021) | 79.6 |
| LipForensics (Haliassos et al., 2021) | 71.4 |
| DCNetwork (Zhou et al., 2023) | 80.5 |
| IID (Huang et al., 2023) | 87.8 |
| DFS (Ye et al., 2024) | 87.3 |
| Clipping (Khan and Dang-Nguyen, 2024) | 91.1 |
| Ours | 93.5 |
Models trained on FF++ and tested on Celeb-DF. Metrics are AUC (%). The bold values indicate the best performance during a comparison.