Medical Visual Question Answering on MedXpertQA OOD (test)
0.184ECEOurs
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| OursModel Architecture=MedGemma-4B-IT, Training Pathway=OmniMedVQA-trained2026.06 | 0.184 | 0.257 | 0.514 | 23.4 | |
| OursTraining=PMC-trained, Pathway=PMC-VQA-trained2026.06 | 0.205 | 0.248 | 0.483 | 22.9 | |
| OursBackbone=Qwen2-VL-7B-Instruct, Trained on=OmniMedVQA2026.06 | 0.314 | 0.278 | 0.487 | 19.4 | |
| SteerConfBackbone=Qwen2-VL-7B-Instruct, Trained on=OmniMedVQA2026.06 | 0.351 | 0.315 | 0.52 | 19.6 | |
| ConfTunerBackbone=Qwen2-VL-7B-Instruct, Trained on=OmniMedVQA2026.06 | 0.36 | 0.369 | 0.504 | 20.9 | |
| SteerConfPathway=PMC-VQA-trained2026.06 | 0.487 | 0.448 | 0.501 | 23.8 | |
| SteerConfModel Architecture=MedGemma-4B-IT, Training Pathway=OmniMedVQA-trained2026.06 | 0.487 | 0.448 | 0.501 | 23.8 | |
| ConfTunerModel Architecture=MedGemma-4B-IT, Training Pathway=OmniMedVQA-trained2026.06 | 0.561 | 0.494 | 0.507 | 21.8 | |
| Top-K SamplingBackbone=Qwen2-VL-7B-Instruct, Trained on=OmniMedVQA2026.06 | 0.569 | 0.491 | 0.491 | 20.6 | |
| Base ModelPathway=PMC-VQA-trained2026.06 | 0.656 | 0.617 | 0.486 | 23.8 | |
| Base ModelModel Architecture=MedGemma-4B-IT, Training Pathway=OmniMedVQA-trained2026.06 | 0.656 | 0.617 | 0.486 | 23.8 | |
| ConfTunerPathway=PMC-VQA-trained2026.06 | 0.678 | 0.641 | 0.478 | 22.6 | |
| Top-K SamplingPathway=PMC-VQA-trained2026.06 | 0.686 | 0.628 | 0.471 | 18.4 | |
| Top-K SamplingModel Architecture=MedGemma-4B-IT, Training Pathway=OmniMedVQA-trained2026.06 | 0.686 | 0.628 | 0.471 | 18.4 | |
| Base ModelBackbone=Qwen2-VL-7B-Instruct, Trained on=OmniMedVQA2026.06 | 0.79 | 0.786 | 0.501 | 19.8 |