Visual Question Answering on PathVQA (open-ended)
4.88Exact Match (EM)BDG-Classic
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| BDG-ClassicModel=Qwen3-VL 8B2026.05 | 4.88 | 10.06 | 40.61 | 11.84 | — | |
| BDG-WModel=Qwen3-VL 8B2026.05 | 4.88 | 10.18 | 40.69 | 12.14 | — | |
| SCDModel=Qwen3-VL 8B2026.05 | 3.38 | 7.57 | 38.99 | 9.45 | — | |
| VerifierModel=Qwen3-VL 8B2026.05 | 3.18 | 8.85 | 40.01 | 11.94 | — | |
| BDG-ClassicModel=Qwen3-VL 4B2026.05 | 2.99 | 6.88 | 38.97 | 11.24 | — | |
| BDG-WModel=Qwen3-VL 4B2026.05 | 2.99 | 6.74 | 38.79 | 11.14 | — | |
| GreedyModel=Qwen3-VL 8B2026.05 | 2.99 | 7.17 | 39.94 | 10.95 | — | |
| VerifierModel=Qwen3-VL 4B2026.05 | 2.79 | 7.72 | 39.64 | 13.33 | — | |
| VerifierModel=Qwen3-VL 2B2026.05 | 2.59 | 5.71 | 37.37 | 6.57 | — | |
| BDG-WModel=Qwen3-VL 2B2026.05 | 2.49 | 5.05 | 36.98 | 7.17 | — | |
| GreedyModel=Qwen3-VL 4B2026.05 | 2.49 | 6.65 | 37.59 | 7.96 | — | |
| BDG-ClassicModel=MedGemma 4B2026.05 | 2.39 | 5.16 | 36.75 | 26.17 | — | |
| BDG-WModel=MedGemma 4B2026.05 | 2.39 | 5.17 | 36.86 | 25.87 | — | |
| BDG-ClassicModel=Qwen3-VL 2B2026.05 | 2.29 | 4.77 | 36.88 | 6.93 | — | |
| SCDModel=Qwen3-VL 4B2026.05 | 2.09 | 5.89 | 37.61 | 10.35 | — | |
| GreedyModel=MedGemma 4B2026.05 | 1.99 | 3.21 | 36.55 | 23.88 | — | |
| GreedyModel=Qwen3-VL 2B2026.05 | 1.49 | 3.79 | 36.32 | 4.98 | — | |
| VerifierModel=MedGemma 4B2026.05 | 0.9 | 2.96 | 35.76 | 24.78 | — | |
| SCDModel=Qwen3-VL 2B2026.05 | 0.8 | 3.02 | 35.38 | 2.79 | — | |
| SCDModel=MedGemma 4B2026.05 | 0.5 | 2.27 | 36.37 | 26.77 | — | |
| GreedyModel=Gemma-3 4B2026.05 | 0 | 2.2 | 35.34 | 18.91 | — | |
| SCDModel=Gemma-3 4B2026.05 | 0 | 2.5 | 34.99 | 20.1 | — | |
| VerifierModel=Gemma-3 4B2026.05 | 0 | 2.19 | 35.06 | 22.99 | — | |
| BDG-ClassicModel=Gemma-3 4B2026.05 | 0 | 2.77 | 35.42 | 24.78 | — | |
| BDG-WModel=Gemma-3 4B2026.05 | 0 | 2.89 | 35.41 | 23.89 | — | |
| ExGra-MedFine-tuning protocol=LoRA, LoRA rank=322026.06 | — | — | — | — | 37.9 | |
| HuatuoGPT-Vision 7BFine-tuning protocol=LoRA, LoRA rank=322026.06 | — | — | — | — | 37.1 | |
| LLaVA-Med 1.5Fine-tuning protocol=LoRA, LoRA rank=322026.06 | — | — | — | — | 38 | |
| MEDSIGHTFine-tuning protocol=LoRA, LoRA rank=322026.06 | — | — | — | — | 38.1 |