Forensic Attribution on FakeBench MCQ
79.47Yes/No AccuracyGPT-4V (Teacher)
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| GPT-4V (Teacher)2025.03 | 79.47 | 68.8 | 55.87 | 62.4 | 70.7 | 61.5 | 67.6 | 66.6 | |
| FakeScope-LBase Model=LLaVA-1.52025.03 | 79.47 | 78 | 78.27 | 75.2 | 82.8 | 73.7 | 76.7 | 77.74 | |
| FakeScope-MBase Model=mPLUG-Owl22025.03 | 75.47 | 79.73 | 78.53 | 80.27 | 76.8 | 80.5 | 78.2 | 78.5 | |
| mPLUG-Owl22025.03 | 70.53 | 66.93 | 62.27 | 64.13 | 64.8 | 66.3 | 66.8 | 65.97 | |
| GeminiPro2025.03 | 70.37 | 66.2 | 52.13 | 70.8 | 65.2 | 62.6 | 66.8 | 64.88 | |
| InternLM-XC.2-vl2025.03 | 69.73 | 68.13 | 66.27 | 59.47 | 66.3 | 65.5 | 65.9 | 65.9 | |
| Claude3 Sonnet2025.03 | 68.27 | 67.73 | 66.8 | 65.2 | 68.3 | 68.8 | 63.9 | 67 | |
| Qwen-VL2025.03 | 64.13 | 64.93 | 60.93 | 60.67 | 62.3 | 61.8 | 63.9 | 62.67 | |
| LLaVA-v1.52025.03 | 60.4 | 62.27 | 63.07 | 61.6 | 63.8 | 62.6 | 59.11 | 61.83 | |
| InstructBLIP2025.03 | 56.27 | 56 | 54.93 | 53.6 | 57.7 | 58.4 | 49.5 | 55.2 | |
| Random guess2025.03 | 50.13 | 37.6 | 24.27 | 26.93 | 34.5 | 37.8 | 31.9 | 34.73 | |
| IDEFICS-Instruct2025.03 | 48.67 | 49.6 | 52.27 | 53.47 | 50.2 | 49.7 | 53.1 | 51 |