Multimodal Misinformation Detection on MMFakeBench
82AccuracyFact-checking (ours)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Fact-checking (ours)Variant=Evidence-centric fact-checking system2026.02 | 82 | 82 | 81 | 81 | |
| FatFormerType=image-only deepfake detector2026.02 | 79 | 61 | 53 | 53 | |
| Fact-checking + deepfake detectorVariant=Hybrid (detector-integrated)2026.02 | 76 | 74 | 74 | 74 | |
| GRAMNetType=image-only deepfake detector2026.02 | 74 | 48 | 49 | 47 | |
| BNext-MType=image-only deepfake detector2026.02 | 74 | 58 | 51 | 53 | |
| NPRType=image-only deepfake detector2026.02 | 64 | 52 | 53 | 52 | |
| COREQwenevaluation_setting=cross-dataset zero-shot, training_samples=3k mixed2026.06 | 62.6 | — | — | — | |
| HAMMERevaluation_setting=cross-dataset zero-shot, training_samples=3k mixed2026.06 | 54.1 | — | — | — | |
| FKA-Owlevaluation_setting=cross-dataset zero-shot, training_samples=3k mixed2026.06 | 53.3 | — | — | — | |
| AMDevaluation_setting=cross-dataset zero-shot, training_samples=3k mixed2026.06 | 46.1 | — | — | — | |
| ProDetType=image-only deepfake detector2026.02 | 26 | 54 | 52 | 25 |