Table 6 Cross-manipulation evaluation on Celeb-DFv1 and Celeb-DFv2 (AUC(%)) after training on FFpp dataset
From: Assessment framework for deepfake detection in real-world situations
Deepfake Detector | Augmentation Method | Â | FFpp | Celeb-DFv1 | Celeb-DFv2 |
|---|---|---|---|---|---|
Capsule | No Aug | Â | 99.20 | 43.36 | 54.39 |
Augmix | Â | 98.66 | 53.45 | 58.65 | |
DAug | Â | 93.51 | 71.35 | 68.39 | |
SDAug | Â | 97.82 | 74.84 | 71.86 | |
XceptionNet | No Aug | Â | 99.56 | 39.35 | 50.00 |
Augmix | Â | 99.15 | 50.27 | 53.04 | |
DAug | Â | 78.64 | 64.79 | 62.81 | |
SDAug | Â | 98.44 | 80.67 | 73.88 |