Visual Misinterpretation Hallucination Evaluation on CXR-VisHal (test)
89.3Accuracy (A)MedQwen
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| MedQwenevaluation_mode=close-ended2026.04 | 89.3 | 70.2 | 83.9 | 91.6 | 82.5 | |
| GPT-4oevaluation_mode=close-ended2026.04 | 88 | 59.5 | 78.8 | 92.1 | 79.4 | |
| LLaVA-Med-1.5evaluation_mode=close-ended2026.04 | 84 | 49.4 | 65.1 | 84.5 | 68.4 | |
| LLaVA-NeXT 7Bevaluation_mode=close-ended2026.04 | 81.7 | 43 | 47.4 | 36.2 | 51.8 | |
| CheXagentevaluation_mode=close-ended2026.04 | 78.2 | 57.6 | 73.9 | 85.1 | 73.9 | |
| LLaVA-NeXT 13Bevaluation_mode=close-ended2026.04 | 77.6 | 39.1 | 48.6 | 56.3 | 53.4 | |
| LLaVA-Medevaluation_mode=close-ended2026.04 | 69.8 | 45.2 | 72.5 | 80 | 69.8 | |
| LLM-CXRevaluation_mode=close-ended2026.04 | 68.1 | 50.4 | 74.3 | 40.3 | 67.5 | |
| Med-Flamingoevaluation_mode=close-ended2026.04 | 36.1 | 32.4 | 57.6 | 33.2 | 48.9 | |
| MiniGPT-4evaluation_mode=close-ended2026.04 | 34.1 | 30.1 | 57.3 | 35.4 | 48.3 |