Deepfake Detection on FF++ (test)
99.96AUCLAA-Net
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LAA-NetEvaluation Protocol=In-distribution2025.11 | 99.96 | — | — | |
| MADDTraining Set=FF++2022.03 | 99.8 | — | — | |
| ID-unaware Deepfake Detection ModelBackbone=Efficient-b42022.10 | 99.79 | — | — | |
| ID-unaware Deepfake Detection ModelBackbone=Efficient-b32022.10 | 99.78 | — | — | |
| XceptionTraining Set=FF++2022.03 | 99.7 | — | — | |
| ID-unaware Deepfake Detection ModelBackbone=ResNet-342022.10 | 99.7 | — | — | |
| SBIBackbone=Efficient-b42022.10 | 99.64 | — | — | |
| MATBackbone=Efficient-b42022.10 | 99.61 | — | — | |
| XceptionBackbone=Xception2022.10 | 99.58 | — | — | |
| MMMSBackbone=Transformer2022.10 | 99.5 | — | — | |
| FauxNetEvaluation Protocol=In-distribution2025.11 | 99.38 | 96 | — | |
| MARE2026.01 | 99.28 | 97.5 | — | |
| M2F2Venue=CVPR20252026.01 | 99.25 | 97.38 | — | |
| Face-x-rayBackbone=HRNet2022.10 | 99.17 | — | — | |
| PCL+I2GBackbone=ResNet-342022.10 | 99.11 | — | — | |
| SLADDTraining Set=FF++2022.03 | 98.4 | — | — | |
| SLADDBackbone=Xception2022.10 | 98.4 | — | — | |
| ExpDVenue=AAAI20242026.01 | 98.4 | 95.6 | — | |
| VideoMAEEvaluation Protocol=In-distribution2025.11 | 98.3 | 94.57 | — | |
| F3-NetBackbone=Xception2022.10 | 98.1 | — | — | |
| SPSLBackbone=Xception2022.10 | 96.91 | — | — | |
| SPSLTraining Set=FF++2022.03 | 96.9 | — | — | |
| SMILTraining Set=FF++2022.03 | 96.8 | — | — | |
| CapsuleTraining Set=FF++2022.03 | 96.6 | — | — | |
| EfficientNetVenue=ICML20192026.01 | 94.18 | 91.19 | — | |
| EfficientNet B4Evaluation Protocol=In-distribution2025.11 | 93.82 | — | — | |
| Two-branchTraining Set=FF++2022.03 | 93.2 | — | — | |
| Two-branchBackbone=DenseNet2022.10 | 93.2 | — | — | |
| DSP-FWATraining Set=FF++2022.03 | 93 | — | — | |
| DSP-FWABackbone=ResNet-502022.10 | 93 | — | — | |
| FauxNetEvaluation Protocol=Zero-shot, Training Set=Authentica-Vox2025.11 | 91.43 | 84.43 | — | |
| SFICVenue=TIFS20242026.01 | 90.3 | 90.24 | — | |
| Meso4Training Set=FF++2022.03 | 84.7 | — | — | |
| MesoInception4Training Set=FF++2022.03 | 83 | — | — | |
| ID-RevealModality=V, w/ Ref.=✓, Train set=VoxCeleb22025.10 | 81.28 | — | 88.19 | |
| FWATraining Set=FF++2022.03 | 80.1 | — | — | |
| RefereeModality=AV, w/ Ref.=✓, Train set=FakeAVCeleb2025.10 | 79.78 | — | 91 | |
| Multi-task2022.10 | 76.3 | — | — | |
| Two-streamTraining Set=FF++2022.03 | 70.1 | — | — | |
| AVADModality=AV, w/ Ref.=✗, Train set=LRS2/LRS32025.10 | 67.01 | — | 88.7 | |
| VA-MLPTraining Set=FF++2022.03 | 66.4 | — | — | |
| XceptionModality=V, w/ Ref.=✗, Train set=FakeAVCeleb2025.10 | 59.48 | — | 76.6 | |
| POI-ForensicsModality=AV, w/ Ref.=✓, Train set=VoxCeleb22025.10 | 51.33 | — | 84.75 | |
| HeadposeTraining Set=FF++2022.03 | 47.3 | — | — | |
| CNN1# Params=537,9942026.02 | — | 82.44 | — | |
| CNN2# Params=108,6182026.02 | — | 79.85 | — | |
| Mapping CNN (CNN1)# Params=1024, Training Strategy=Full Network training, Base Architecture=CNN12026.02 | — | 81.11 | — | |
| Mapping CNN (CNN1)# Params=2048, Training Strategy=Full Network training, Base Architecture=CNN12026.02 | — | 85.23 | — | |
| Mapping CNN (CNN1)# Params=1956, Training Strategy=Layer-wise training, Base Architecture=CNN12026.02 | — | 86.23 | — | |
| Mapping CNN (CNN1)# Params=2792, Training Strategy=Layer-wise training, Base Architecture=CNN12026.02 | — | 88.05 | — | |
| Mapping CNN (CNN2)# Params=1024, Training Strategy=Full Network training, Base Architecture=CNN22026.02 | — | 82.78 | — | |
| Mapping CNN (CNN2)# Params=2048, Training Strategy=Full Network training, Base Architecture=CNN22026.02 | — | 84.09 | — | |
| Mapping CNN (CNN2)# Params=1872, Training Strategy=Layer-wise training, Base Architecture=CNN22026.02 | — | 83.1 | — | |
| Mapping CNN (CNN2)# Params=2688, Training Strategy=Layer-wise training, Base Architecture=CNN22026.02 | — | 86.28 | — | |
| Mapping Networks# Params=2048, Layers=All2026.02 | — | 91.02 | — | |
| Mapping Networks# Params=1024, Layers=L-4, FC2026.02 | — | 89.23 | — | |
| ResNet50# Params=25M, Layers=All2026.02 | — | 91.78 | — | |
| ResNet50# Params=17M, Layers=L-4, FC2026.02 | — | 88.03 | — |