Visual Question Answering on MMStar (val)
59.8CP ScoreBM25
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| BM25Base Model=Qwen-VL, Selection Strategy=BM252025.06 | 59.8 | 34.1 | 51.1 | 30.6 | 23.6 | 24.9 | |
| randomBase Model=Qwen-VL, Selection Strategy=random2025.06 | 57.2 | 33.9 | 46.7 | 26.6 | 20.5 | 28.8 | |
| similarityBase Model=Qwen-VL, Selection Strategy=Similarity2025.06 | 57.2 | 31.4 | 47.2 | 32.8 | 26.2 | 27.1 | |
| multi-modal exploration-exploitation reinforcement learning frameworkBase Model=Qwen-VL, Selection Strategy=Reinforcement Learning2025.06 | 56.8 | 36.2 | 46.6 | 34.1 | 24.9 | 29.3 | |
| multi-modal exploration-exploitation reinforcement learning frameworkBase Model=LLaVA, Selection Strategy=Reinforcement Learning2025.06 | 54.6 | 38 | 49.8 | 36.7 | 20.1 | 28.4 | |
| zero-shotBase Model=Qwen-VL, Selection Strategy=zero-shot2025.06 | 52 | 37.6 | 47.6 | 25.3 | 17.9 | 33.2 | |
| similarityBase Model=LLaVA, Selection Strategy=Similarity2025.06 | 48.5 | 28.4 | 43.7 | 33.6 | 23.6 | 27.5 |