Report Generation on IU-Xray
33.1ROUGE-LVALOR
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| VALORBackbone=LLaVA-Medv1.52025.12 | 33.1 | 61.2 | 28.3 | 12.9 | 5.7 | 0.3688 | — | |
| VALORType=General VLM, Zero-shot=true2025.12 | 33.1 | — | — | — | — | 0.3688 | 27.02 | |
| VALORType=Medical VLM, Zero-shot=true2025.12 | 33.1 | — | — | — | — | 0.3688 | 27.02 | |
| LLaVA-Medv1.5 + GRPOTraining Technique=Group Relative Policy Optimization2025.12 | 31.8 | 59.02 | 26.5 | 11.8 | 5.1 | 0.359 | — | |
| MAIRA-2Type=Medical VLM, Zero-shot=true2025.12 | 30.6 | — | — | — | — | 0.3236 | 20.22 | |
| LLaVA-Medv1.5 + MMedPOTraining Technique=MMedPO2025.12 | 29.52 | 55.58 | 23.93 | 10.36 | 4.4 | 0.3416 | — | |
| GPT-4oType=General VLM, Zero-shot=true2025.12 | 18.8 | — | — | — | — | 0.115 | 15.32 | |
| LLaVA-Medv1.5 + DPOTraining Technique=Direct Preference Optimization2025.12 | 12.95 | 41.63 | 15.13 | 5.56 | 2.03 | 0.1713 | — | |
| LLaVA-Medv1.5 + SFTFine-tuning=Supervised Fine-Tuning2025.12 | 10.31 | 38.42 | 13.4 | 4.74 | 1.67 | 0.1095 | — | |
| LLaVA-Medv1.5Model Version=v1.52025.12 | 7.58 | 22.29 | 7.95 | 2.9 | 1.05 | 0.0456 | — |