Video-level Deepfake Detection on DFDC
0.896AUCVLAForge
Evaluation results
| Method | Links | |
|---|---|---|
| VLAForgeVenue=-2026.03 | 0.896 | |
| VLAForgeParam.=3.28 M, Training (ms)=1410.0±2.2, Inference (ms)=41.7±0.92026.03 | 0.896 | |
| M2F2-detDFD (Dedicated Deepfake Detector)=true, MLLM (Multimodal Large Language Model)=true, Training dataset=FF++ c232026.01 | 0.878 | |
| F. Adap.DFD (Dedicated Deepfake Detector)=true, MLLM (Multimodal Large Language Model)=false, Training dataset=FF++ c232026.01 | 0.872 | |
| ForAdaVenue=CVPR’252026.03 | 0.872 | |
| ForAdaParam.=5.7M, Training (ms)=933.3±1.4, Inference (ms)=21.2±0.22026.03 | 0.872 | |
| VLF-FFDDFD (Dedicated Deepfake Detector)=true, MLLM (Multimodal Large Language Model)=true, Training dataset=FF++ c232026.01 | 0.854 | |
| X2-DFDDFD (Dedicated Deepfake Detector)=true, MLLM (Multimodal Large Language Model)=true, Training dataset=FF++ c232026.01 | 0.853 | |
| EffortDFD (Dedicated Deepfake Detector)=true, MLLM (Multimodal Large Language Model)=false, Training dataset=FF++ c232026.01 | 0.843 | |
| Yan et al.Venue=CVPR’252026.03 | 0.843 | |
| EffortVenue=ICML’252026.03 | 0.843 | |
| RLSBIDFD (Dedicated Deepfake Detector)=false, MLLM (Multimodal Large Language Model)=true, Training dataset=FF++ c232026.01 | 0.839 | |
| CDFADFD (Dedicated Deepfake Detector)=true, MLLM (Multimodal Large Language Model)=false, Training dataset=FF++ c232026.01 | 0.83 | |
| CDFAVenue=ECCV’242026.03 | 0.83 | |
| FCGVenue=CVPR’252026.03 | 0.818 | |
| RepDFDVenue=AAAI’252026.03 | 0.81 | |
| RepDFDParam.=0.078M, Training (ms)=1832.4±7.5, Inference (ms)=42.3±0.62026.03 | 0.81 | |
| CFMParam.=25.37M2026.03 | 0.802 | |
| KFDDFD (Dedicated Deepfake Detector)=true, MLLM (Multimodal Large Language Model)=true, Training dataset=FF++ c232026.01 | 0.791 | |
| DCLParam.=19.35M2026.03 | 0.767 | |
| SeeABLEVenue=ICCV’232026.03 | 0.759 | |
| UCFDFD (Dedicated Deepfake Detector)=true, MLLM (Multimodal Large Language Model)=false, Training dataset=FF++ c232026.01 | 0.742 | |
| FTCNParam.=26.6M2026.03 | 0.74 | |
| CADDMVenue=CVPR’232026.03 | 0.739 | |
| SFDGVenue=CVPR’232026.03 | 0.736 | |
| LSDAVenue=CVPR’242026.03 | 0.735 | |
| SBIVenue=CVPR’222026.03 | 0.724 | |
| SBIDFD (Dedicated Deepfake Detector)=true, MLLM (Multimodal Large Language Model)=false, Training dataset=FF++ c232026.01 | 0.717 | |
| ProDetDFD (Dedicated Deepfake Detector)=true, MLLM (Multimodal Large Language Model)=false, Training dataset=FF++ c232026.01 | 0.707 | |
| ProDetVenue=Neurips’242026.03 | 0.707 | |
| LSDADFD (Dedicated Deepfake Detector)=true, MLLM (Multimodal Large Language Model)=false, Training dataset=FF++ c232026.01 | 0.701 | |
| IIDVenue=CVPR’232026.03 | 0.7 | |
| RECCEDFD (Dedicated Deepfake Detector)=true, MLLM (Multimodal Large Language Model)=false, Training dataset=FF++ c232026.01 | 0.696 | |
| TALLVenue=ICCV’232026.03 | 0.693 |