Audio-Visual Deepfake Detection on FakeAVCeleb
99.91AccuracyFGMDF
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FGMDFEvaluation Setting=four-class evaluation2026.04 | 99.91 | 99.97 | |
| Ours (5-class + Semantic)Training Configuration=five-class setting + semantic reinforcement, Evaluation Setting=four-class evaluation2026.04 | 99.58 | 99.96 | |
| ConLLMInitialization weights=GPT Weights2026.01 | 98.75 | 99.98 | |
| Ours (5-class)Training Configuration=five-class setting, Evaluation Setting=four-class evaluation2026.04 | 98.61 | 99.94 | |
| PVASS2026.01 | 95.7 | 97.3 | |
| ConLLMInitialization weights=w/o GPT Weights2026.01 | 95 | 98 | |
| JAVDDEvaluation Setting=four-class evaluation2026.04 | 94.28 | 77.78 | |
| AVAD2026.01 | 94.2 | 94.5 | |
| BA-TFDEvaluation Setting=four-class evaluation2026.04 | 93.77 | 77.57 | |
| MDSEvaluation Setting=four-class evaluation2026.04 | 93.71 | 73.18 | |
| AVoiD-DF2026.01 | 83.7 | 89.2 | |
| MDS2026.01 | 82.8 | 86.5 | |
| JointAV2026.01 | 82.5 | 83.3 | |
| VFD2026.01 | 81.5 | 86.1 | |
| BA-TFD2026.01 | 80.8 | 84.9 | |
| AVFakeNet2026.01 | 78.4 | 83.4 | |
| EmoForen2026.01 | 78.1 | 79.8 |