Medical Report Generation on MMEndo
65.8BLEU-1LDP
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| LDPComponents=LoRA + DPO + PE, Preference Alignment=DPO2025.12 | 65.8 | 59.1 | 61.8 | 63.6 | 52 | 7.2 | |
| Qwen2-VL-7B + LoRA (SFT)PEFT strategy=LoRA, Phase=Supervised Fine-Tuning2025.12 | 64.2 | 58.5 | 60.9 | 63.8 | 52.7 | 6.7 | |
| Qwen2-VL-7B + AdaLoRAPEFT strategy=AdaLoRA2025.12 | 37.4 | 24.3 | 29.6 | 34.7 | 40.2 | 5.8 | |
| Qwen2-VL-7B + PEPrompt Engineering=true2025.12 | 18.2 | 7.2 | 15.7 | 11.7 | 21.3 | 4.5 | |
| Qwen2-VL-7BConfig=Baseline2025.12 | 12.3 | 4.1 | 10.7 | 7.4 | 9.8 | 3.2 |