Table 2 Performance of different models on the FaceForensics++dataset at various compression rates.
From: ID-insensitive deepfake detection model based on multi-attention mechanism
Methods | LQ | HQ | ||
|---|---|---|---|---|
ACC (%) | AUC (%) | ACC (%) | AUC (%) | |
Steg.Features46 | 55.98 | – | 70.97 | – |
LD-CNN47 | 58.69 | – | 78.45 | – |
MesoNet11 | 70.47 | – | 83.10 | – |
Face X-ray48 | – | 61.60 | – | 87.40 |
Xception29 | 86.86 | 89.30 | 95.73 | 96.30 |
Xception-ELA49 | 79.63 | 82.90 | 93.86 | 94.80 |
Xception- PAFilters50 | 87.16 | 90.20 | – | – |
F3-Net51 | 90.43 | 93.30 | 97.52 | 98.10 |
Two Branch31 | – | 86.59 | – | 98.70 |
EfficientNet-B43 | 86.67 | 88.20 | 96.63 | 99.18 |
Multi-Attentional Deepfake Detection8 | 88.69 | 90.40 | 97.60 | 99.29 |
Ours | 91.20 | 94.59 | 98.80 | 99.91 |
