Deepfake Detection on CDFv1, CDFv2, DFD, DFDCP, DFDC (test)
94.4Overall Average ScoreGSD
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| GSD2026.03 | 94.4 | 96.4 | 96.4 | 97.7 | 93.2 | 88.3 | — | |
| ForAdaSource=CVPR’252026.03 | 93.2 | — | 95.7 | 97.1 | 92.9 | 87.2 | — | |
| MASMVenue=-, Training Dataset=FF++ c23, Evaluation Level=Video-level2026.01 | 92.985 | — | — | 98.191 | 91.386 | 87.133 | 95.23 | |
| χ²-DFDSource=NeurIPS’252026.03 | 91.9 | — | 95.5 | 95.7 | 91.2 | 85.3 | — | |
| EffortSource=ICML’25, official_pre-trained_or_reproduction=true2026.03 | 91.8 | 91.8 | 95.6 | 96.5 | 90.9 | 84.3 | — | |
| StAVenue=CVPR 2025, Training Dataset=FF++ c23, Evaluation Level=Video-level2026.01 | 91.6 | — | — | 96.5 | 90.9 | 84.3 | 94.7 | |
| VbSaTSource=CVPR’252026.03 | 91.6 | — | 94.7 | 96.5 | 90.9 | 84.3 | — | |
| DeepShieldSource=ICCV’252026.03 | 91.1 | — | 92.2 | 96.1 | 93.2 | 82.8 | — | |
| CDFASource=ECCV’24, official_pre-trained_or_reproduction=true2026.03 | 90.5 | 92.1 | 93.8 | 95.4 | 88.1 | 83 | — | |
| GSD2026.03 | 90.3 | 91 | 90.6 | 94.3 | 89.9 | 85.5 | — | |
| FakeRadarSource=ICCV’252026.03 | 90.1 | — | 91.7 | 96.2 | 88.5 | 84.1 | — | |
| EffortVenue=ICML 2025, Training Dataset=FF++ c23, Evaluation Level=Video-level, reproduced=true2026.01 | 89.794 | — | — | 95.721 | 85.692 | 85.061 | 92.703 | |
| UDDVenue=AAAI 2025, Training Dataset=FF++ c23, Evaluation Level=Video-level2026.01 | 89.475 | — | — | 95.5 | 88.1 | 81.2 | 93.1 | |
| LaVSource=MM’232026.03 | 89.1 | 84.7 | 92.2 | 96.8 | 89.7 | 82 | — | |
| CDFAVenue=ECCV 2024, Training Dataset=FF++ c23, Evaluation Level=Video-level2026.01 | 88.8 | — | — | 95.4 | 83 | 83 | 93.8 | |
| FCGSource=CVPR’252026.03 | 88.4 | — | 95 | — | — | 81.8 | — | |
| GM-DFSource=MM’252026.03 | 88.2 | 89.3 | 89.2 | 92.8 | 88.9 | 84.7 | — | |
| CLIPSource=ICML’21, official_pre-trained_or_reproduction=true2026.03 | 88 | 89.6 | 88.4 | 92.6 | 86.3 | 83.3 | — | |
| ProDeSource=NeurIPS’24, official_pre-trained_or_reproduction=true2026.03 | 87.4 | 95.7 | 92.6 | 95.4 | 82.8 | 70.7 | — | |
| SBIsSource=CVPR’222026.03 | 85.4 | — | 90.6 | 88.2 | 87.7 | 75.2 | — | |
| FSgDSource=NeurIPS’252026.03 | 85.2 | 90.1 | 86.7 | 82.1 | 81.8 | — | — | |
| UDDSource=AAAI’252026.03 | 84.8 | — | 86.9 | 91 | 85.6 | 75.8 | — | |
| ProDetVenue=NeurIPS 2024, Training Dataset=FF++ c23, Evaluation Level=Video-level, reproduced=true2026.01 | 84.524 | — | — | 90.139 | 82.823 | 72.517 | 92.616 | |
| CLIPSource=ICML’21, official_pre-trained_or_reproduction=true2026.03 | 84 | 86.1 | 83.2 | 87.7 | 82.4 | 80.7 | — | |
| DFFakeSource=NeurIPS’242026.03 | 84 | — | 80.5 | 90.4 | 81 | — | — | |
| CFMSource=TIFS’232026.03 | 83.9 | — | 89.7 | 95.2 | 80.2 | 70.6 | — | |
| ICTSource=CVPR’232026.03 | 83.7 | 81.4 | 85.7 | 84.1 | — | — | — | |
| ExposDeSource=AAAI’24, official_pre-trained_or_reproduction=true2026.03 | 83 | 81.8 | 86.4 | 89.8 | 85.1 | 72.1 | — | |
| FreqDebiasSource=CVPR’252026.03 | 82.9 | 87.5 | 83.6 | 86.8 | 82.4 | 74.1 | — | |
| FreqDebiasEvaluation Setting=Cross-domain, Training Dataset=FF++ (HQ), Evaluation Level=frame-level2025.09 | 82.88 | 87.5 | 83.6 | 86.8 | 82.4 | 74.1 | — | |
| EFNB4 + LSDAMethod Type=Ours, Backbone=EfficientNet, Training Dataset=FF++_c232023.11 | 82.6 | 86.7 | 83 | 88 | 81.5 | 73.6 | — | |
| LSDASource=CVPR’242026.03 | 82.6 | 86.7 | 83 | 88 | 81.5 | 73.6 | — | |
| LSDAEvaluation Setting=Cross-domain, Training Dataset=FF++ (HQ), Evaluation Level=frame-level2025.09 | 82.56 | 86.7 | 83 | 88 | 81.5 | 73.6 | — | |
| ProDeSource=NeurIPS’24, official_pre-trained_or_reproduction=true2026.03 | 82.3 | 90.9 | 84.5 | — | 81.2 | 72.4 | — | |
| LSDAVenue=CVPR 2024, Training Dataset=FF++ c23, Evaluation Level=Video-level2026.01 | 81.725 | — | — | 88.1 | 81.2 | 70.1 | 87.5 | |
| UCFSource=ICCV’23, official_pre-trained_or_reproduction=true2026.03 | 80.9 | 77.9 | 82.4 | 87.8 | 75.9 | 80.5 | — | |
| DAIDSource=NeurIPS’252026.03 | 80.8 | — | 84.4 | 91.2 | — | 66.9 | — | |
| UCFVenue=ICCV 2023, Training Dataset=FF++ c23, Evaluation Level=Video-level2026.01 | 80.4 | — | — | 86.7 | 77 | 74.2 | 83.7 | |
| IIDSource=CVPR’232026.03 | 79.2 | — | 83.8 | 93.9 | 68.9 | 70 | — | |
| F3NetSource=ECCV’202026.03 | 77.5 | — | 78.9 | 84.4 | 74.9 | 71.8 | — | |
| SPSLEvaluation Setting=Cross-domain, Training Dataset=FF++ (HQ), Evaluation Level=frame-level2025.09 | 76.74 | 81.5 | 76.5 | 81.2 | 74.1 | 70.4 | — | |
| SPSLMethod Type=Frequency, Backbone=Xception, Training Dataset=FF++_c232023.11 | 76.7 | 81.5 | 76.5 | 81.2 | 74.1 | 70.4 | — | |
| SLADDEvaluation Setting=Cross-domain, Training Dataset=FF++ (HQ), Evaluation Level=frame-level2025.09 | 76.42 | 80.2 | 74 | 80.9 | 75.3 | 71.7 | — | |
| UCFEvaluation Setting=Cross-domain, Training Dataset=FF++ (HQ), Evaluation Level=frame-level2025.09 | 76.34 | 77.9 | 75.3 | 80.7 | 75.9 | 71.9 | — | |
| UCFMethod Type=Spatial, Backbone=Xception, Training Dataset=FF++_c232023.11 | 76.3 | 77.9 | 75.3 | 80.7 | 75.9 | 71.9 | — | |
| SRMEvaluation Setting=Cross-domain, Training Dataset=FF++ (HQ), Evaluation Level=frame-level2025.09 | 76.02 | 79.3 | 75.5 | 81.2 | 74.1 | 70 | — | |
| IIDVenue=CVPR 2023, Training Dataset=FF++ c23, Evaluation Level=Video-level, reproduced=true2026.01 | 76.009 | — | — | 82.761 | 71.688 | 70.866 | 78.723 | |
| SRMMethod Type=Frequency, Backbone=Xception, Training Dataset=FF++_c232023.11 | 76 | 79.3 | 75.5 | 81.2 | 74.1 | 70 | — | |
| EfficientB4Method Type=Naive, Backbone=EfficientNet, Training Dataset=FF++_c232023.11 | 75.6 | 79.1 | 74.9 | 81.5 | 72.8 | 69.6 | — | |
| XceptionEvaluation Setting=Cross-domain, Training Dataset=FF++ (HQ), Evaluation Level=frame-level2025.09 | 75.54 | 77.9 | 73.7 | 81.6 | 73.7 | 70.8 | — | |
| IIDEvaluation Setting=Cross-domain, Training Dataset=FF++ (HQ), Evaluation Level=frame-level2025.09 | 75.54 | 75.8 | 76.9 | 79.3 | 76.2 | 69.5 | — | |
| FFDEvaluation Setting=Cross-domain, Training Dataset=FF++ (HQ), Evaluation Level=frame-level2025.09 | 75.52 | 78.4 | 74.4 | 80.2 | 74.3 | 70.3 | — | |
| XceptionMethod Type=Naive, Backbone=Xception, Training Dataset=FF++_c232023.11 | 75.5 | 77.9 | 73.7 | 81.6 | 73.7 | 70.8 | — | |
| FFDMethod Type=Spatial, Backbone=Xception, Training Dataset=FF++_c232023.11 | 75.5 | 78.4 | 74.4 | 80.2 | 74.3 | 70.3 | — | |
| RECCEEvaluation Setting=Cross-domain, Training Dataset=FF++ (HQ), Evaluation Level=frame-level2025.09 | 75.34 | 76.8 | 73.2 | 81.2 | 74.2 | 71.3 | — | |
| COREMethod Type=Spatial, Backbone=Xception, Training Dataset=FF++_c232023.11 | 75.3 | 78 | 74.3 | 80.2 | 73.4 | 70.5 | — | |
| COREEvaluation Setting=Cross-domain, Training Dataset=FF++ (HQ), Evaluation Level=frame-level2025.09 | 75.28 | 78 | 74.3 | 80.2 | 73.4 | 70.5 | — | |
| RecceMethod Type=Spatial, Backbone=Designed, Training Dataset=FF++_c232023.11 | 75.2 | 76.8 | 73.2 | 81.2 | 73.4 | 71.3 | — | |
| F3NetEvaluation Setting=Cross-domain, Training Dataset=FF++ (HQ), Evaluation Level=frame-level2025.09 | 74.94 | 77.7 | 73.5 | 79.8 | 73.5 | 70.2 | — | |
| F3NetMethod Type=Frequency, Backbone=Xception, Training Dataset=FF++_c232023.11 | 74.9 | 77.7 | 73.5 | 79.8 | 73.5 | 70.2 | — | |
| Efficient-b4Source=ICML’19, official_pre-trained_or_reproduction=true2026.03 | 73.9 | 70.5 | 74.9 | 81.5 | 72.8 | 69.6 | — | |
| RECCESource=CVPR’222026.03 | 73.2 | — | 73.2 | 74.2 | 74.2 | 71.3 | — | |
| AltFreezingVenue=CVPR 2023, Training Dataset=FF++ c23, Evaluation Level=Video-level, reproduced=true2026.01 | 70.672 | — | — | 71.838 | 66.689 | 64.015 | 80.146 | |
| CapsuleEvaluation Setting=Cross-domain, Training Dataset=FF++ (HQ), Evaluation Level=frame-level2025.09 | 70.52 | 79.1 | 74.7 | 68.4 | 65.7 | 64.7 | — | |
| CapsuleNetMethod Type=Spatial, Backbone=Capsule, Training Dataset=FF++_c232023.11 | 70.5 | 79.1 | 74.7 | 68.4 | 65.7 | 64.7 | — | |
| X-rayEvaluation Setting=Cross-domain, Training Dataset=FF++ (HQ), Evaluation Level=frame-level2025.09 | 69.62 | 70.9 | 67.9 | 76.6 | 69.4 | 63.3 | — | |
| Face X-rayMethod Type=Spatial, Backbone=HRNet, Training Dataset=FF++_c232023.11 | 69.6 | 70.9 | 67.9 | 76.6 | 69.4 | 63.3 | — | |
| Face X-RaySource=CVPR’202026.03 | 69.6 | 70.9 | 67.9 | 76.7 | 69.4 | 63.3 | — | |
| FWAMethod Type=Spatial, Backbone=Xception, Training Dataset=FF++_c232023.11 | 69 | 79 | 66.8 | 74 | 63.8 | 61.3 | — | |
| MesoIncepMethod Type=Naive, Backbone=MesoNet, Training Dataset=FF++_c232023.11 | 68.4 | 73.7 | 69.7 | 60.7 | 75.6 | 62.3 | — | |
| CNN-AugMethod Type=Naive, Backbone=ResNet, Training Dataset=FF++_c232023.11 | 66.9 | 74.2 | 70.3 | 64.6 | 61.7 | 63.6 | — | |
| Meso4Method Type=Naive, Backbone=MesoNet, Training Dataset=FF++_c232023.11 | 61 | 73.6 | 60.9 | 54.8 | 59.9 | 55.6 | — | |
| Meso4Evaluation Setting=Cross-domain, Training Dataset=FF++ (HQ), Evaluation Level=frame-level2025.09 | 60.96 | 73.6 | 60.9 | 54.8 | 59.9 | 55.6 | — | |
| TALLVenue=ICCV 2023, Training Dataset=FF++ c23, Evaluation Level=Video-level, reproduced=true2026.01 | 55.786 | — | — | 50.843 | 56.148 | 52.201 | 63.951 |