Medical Visual Question Answering on VQA-Med
73.3AccuracyRform + Rexact + RDTW
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Rform + Rexact + RDTWReward functions=Rform + Rexact + RDTW2026.04 | 73.3 | — | — | |
| Rform + RexactReward functions=Rform + Rexact2026.04 | 71.7 | — | — | |
| RformReward functions=Rform2026.04 | 68.8 | — | — | |
| SFTTraining configuration=SFT2026.04 | 68.1 | — | — | |
| Ours (Med-Evo)base model=MedVLM-R12026.03 | 48.62 | 10.18 | 23.42 | |
| EN-INFbase model=MedVLM-R12026.03 | 46.79 | 10.18 | 23.02 | |
| TTRVbase model=MedVLM-R12026.03 | 44.95 | 10.18 | 21.83 | |
| Base Modelbase model=MedVLM-R12026.03 | 44.04 | 9.58 | 23.02 | |
| TTRLbase model=MedVLM-R12026.03 | 38.53 | 6.59 | 20.23 |