General VQA on MMStar (test)
71.3Overall AccuracyOurs (CDRL + CA-TTS)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Ours (CDRL + CA-TTS)Framework Type=Training-Based, Base Model=Qwen2.5-VL-7B2026.03 | 71.3 | 67.8 | 77.2 | |
| Majority VotingFramework Type=Training-Free, Base Model=Qwen2.5-VL-7B2026.03 | 69 | 63.6 | 77 | |
| We-ThinkFramework Type=Training-Based, Base Model=Qwen2.5-VL-7B2026.03 | 65.1 | 62.6 | 70.1 | |
| R1-OnevisionFramework Type=Training-Based, Base Model=Qwen2.5-VL-7B2026.03 | 63.7 | 60.8 | 66.9 | |
| VL-RethinkerFramework Type=Training-Based, Base Model=Qwen2.5-VL-7B2026.03 | 63.4 | 60.7 | 66.5 | |
| DreamPRMFramework Type=Training-Based, Base Model=InternVL-2.5-8B2026.03 | 62.3 | — | — | |
| DeepconfFramework Type=Training-Free, Base Model=Qwen2.5-VL-7B2026.03 | 61.1 | 58.2 | 71 | |
| Pass@1Framework Type=Training-Free, Base Model=Qwen2.5-VL-7B2026.03 | 60.2 | 58.4 | 68.2 |