Multi-modal Forgery Detection and Grounding on MDSM Guardian
92.6AccuracyFKA-Owl
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| FKA-OwlSetting=Tr. on Guardian2025.05 | 92.6 | 78.24 | 71.04 | |
| AMDSetting=Tr. on Guardian2025.05 | 91.43 | 80.85 | 85.09 | |
| ViLTSetting=Tr. on Guardian2025.05 | 85.29 | 67.34 | 41.8 | |
| HAMMER++Setting=Tr. on Guardian2025.05 | 80.95 | 59.92 | 64.67 | |
| AMDSetting=Tr. on NYT2025.05 | 80.21 | 64 | 62.51 | |
| HAMMERSetting=Tr. on Guardian2025.05 | 78.5 | 52.4 | 51.53 | |
| FKA-OwlSetting=Tr. on NYT2025.05 | 77.2 | 46.88 | 43.67 | |
| ViLTSetting=Tr. on NYT2025.05 | 72.18 | 31.76 | 20.21 | |
| HAMMERSetting=Tr. on NYT2025.05 | 68.23 | 40.1 | 21.56 | |
| HAMMER++Setting=Tr. on NYT2025.05 | 66.99 | 38.07 | 17.34 | |
| Gemini-2.0Setting=Zero-Shot2025.05 | 41.26 | 24.37 | 1.12 | |
| Qwen2-a22bSetting=Zero-Shot2025.05 | 38.12 | 27.41 | 1.19 | |
| GPT-4oSetting=Zero-Shot2025.05 | 35.68 | 29.49 | 1.23 | |
| Qwen2.5-VLSetting=Zero-Shot2025.05 | 35.18 | 25.7 | 0 |