Visual Question Answering on InfoSeek (Acc, SR img, SR txt)
34.6AccuracyMMSearch-R1
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MMSearch-R1Model Type=Trained Base Model2026.03 | 34.6 | 56.2 | 36.8 | |
| TAMerging Strategy=Task Arithmetic2026.03 | 22.85 | 35.55 | 38.8 | |
| Qwen-VL-RAGWorkflow=RAG-based2026.03 | 21.3 | 100 | 100 | |
| OBMMerging Strategy=Optimal Brain Merging2026.03 | 19.65 | 31.7 | 46.15 | |
| Qwen-VL-R1Workflow=Search-augmented2026.03 | 19.45 | 11.85 | 16.5 | |
| Qwen-VL-DAWorkflow=Direct Answering2026.03 | 17.9 | 0 | 0 | |
| TIESMerging Strategy=TIES-Merging2026.03 | 15.1 | 2.45 | 37 | |
| DAREMerging Strategy=DARE2026.03 | 10.65 | 12.5 | 35.75 |