Face Forgery Detection on FaceForensics++ (test)
100AUC (DF)SOLA -weakly sup
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SOLA -weakly supBackbone=ResNet18, Train Set=DF2022.09 | 100 | 97.29 | 63.59 | 98.45 | 89.83 | |
| SOLA -supBackbone=ResNet18, Train Set=DF2022.09 | 100 | 96.95 | 69.72 | 98.48 | 91.28 | |
| AV DFD2022.01 | 100 | 99.8 | 90.5 | 98.3 | — | |
| RealForensics2022.01 | 100 | 99.7 | 97.1 | 99.2 | — | |
| SOLA -weakly supBackbone=ResNet18, Train Set=NT2022.09 | 99.95 | 94.83 | 57.32 | 99.82 | 87.97 | |
| FTCN2022.01 | 99.9 | 99.7 | 99.9 | 99.2 | — | |
| FTCNNumber of parameters=26.6M, Pre-training=No, Extra data usage=No, Evaluation level=video-level2021.08 | 99.9 | 99.7 | 99.9 | 99.2 | 99.7 | |
| SOLA -supBackbone=ResNet18, Train Set=F2F2022.09 | 99.73 | 99.56 | 93.5 | 96.02 | 97.2 | |
| LipForensics2022.01 | 99.7 | 99.7 | 90.1 | 99.1 | — | |
| LipForensicsNumber of parameters=36.0M, Pre-training=Yes, Extra data usage=No, Evaluation level=video-level2021.08 | 99.7 | 99.7 | 90.1 | 99.1 | 97.1 | |
| SOLA -supBackbone=ResNet18, Train Set=NT2022.09 | 99.64 | 97.69 | 90.2 | 99.76 | 96.82 | |
| SOLA -weakly supBackbone=ResNet18, Train Set=F2F2022.09 | 99.61 | 99.67 | 84.24 | 97.48 | 95.25 | |
| Face X-ray2022.01 | 99.5 | 94.5 | 93.2 | 92.5 | — | |
| Face X-rayNumber of parameters=65.8M, Pre-training=No, Extra data usage=Yes, Evaluation level=video-level2021.08 | 99.5 | 94.5 | 93.2 | 92.5 | 94.9 | |
| XceptionBackbone=Xception, Train Set=DF2022.09 | 99.38 | 75.05 | 49.13 | 80.39 | 75.99 | |
| Face X-rayBackbone=HRNet, Train Set=DF2022.09 | 99.17 | 94.14 | 75.34 | 93.85 | 90.63 | |
| Face X-rayBackbone=HRNet, Train Set=NT2022.09 | 99.14 | 98.43 | 70.56 | 98.93 | 91.77 | |
| SOLA -supBackbone=ResNet18, Train Set=FS2022.09 | 99.11 | 98.13 | 99.98 | 92.07 | 97.32 | |
| CSN2022.01 | 98.8 | 98.7 | 87.9 | 88.6 | — | |
| Face X-rayBackbone=HRNet, Train Set=F2F2022.09 | 98.52 | 99.06 | 72.69 | 91.49 | 90.44 | |
| CNN-GRU2022.01 | 97.6 | 85.8 | 47.6 | 86.6 | — | |
| CNN-GRUEvaluation level=video-level2021.08 | 97.6 | 85.8 | 47.6 | 86.6 | 79.4 | |
| Patch-based2022.01 | 94 | 87.3 | 60.5 | 84.8 | — | |
| PatchForensicsEvaluation level=video-level2021.08 | 94 | 87.3 | 60.5 | 84.8 | 81.7 | |
| Xception2022.01 | 93.9 | 86.8 | 51.2 | 79.7 | — | |
| XceptionEvaluation level=video-level2021.08 | 93.9 | 86.8 | 51.2 | 79.7 | 77.9 | |
| Face X-rayBackbone=HRNet, Train Set=FS2022.09 | 93.77 | 92.29 | 99.2 | 86.63 | 92.97 | |
| SOLA -weakly supBackbone=ResNet18, Train Set=FS2022.09 | 93.18 | 97.59 | 100 | 94.93 | 96.43 | |
| XceptionBackbone=Xception, Train Set=NT2022.09 | 93.09 | 84.82 | 47.98 | 99.5 | 81.35 | |
| LipForensics-ScratchNumber of parameters=36.0M, Pre-training=No, Extra data usage=No, Evaluation level=video-level2021.08 | 93 | 98.8 | 56.7 | 98.3 | 86.7 | |
| XceptionBackbone=Xception, Train Set=F2F2022.09 | 87.56 | 99.53 | 65.23 | 65.9 | 79.56 | |
| CNN-aug2022.01 | 87.5 | 80.1 | 56.3 | 67.8 | — | |
| CNN-augEvaluation level=video-level2021.08 | 87.5 | 80.1 | 56.3 | 67.8 | 72.9 | |
| XceptionBackbone=Xception, Train Set=FS2022.09 | 70.12 | 61.7 | 99.36 | 68.71 | 74.97 |