Failure Discovery on COCO
88.37QVRExpert Iter.
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Expert Iter.Model=Qwen-2.5-VL-3B2026.04 | 88.37 | 38.36 | 53.42 | 19 | 94 | 3.1 | |
| RL+SFTModel=Qwen-2.5-VL-3B2026.04 | 86.7 | 50.73 | 85.41 | 57 | 122 | 3.27 | |
| SFTModel=Qwen-2.5-VL-3B2026.04 | 83.86 | 37.02 | 43.8 | 23 | 89 | 3.13 | |
| Zero ShotModel=Qwen-2.5-VL-3B2026.04 | 83.73 | 32.2 | 56.27 | 30 | 87 | 3.06 | |
| RLModel=Qwen-2.5-VL-3B2026.04 | 83.44 | 47.58 | 64.96 | 53 | 131 | 3.3 | |
| ConMeModel=Qwen-2.5-VL-3B2026.04 | 76.63 | 39.89 | 61.78 | 30 | 93 | 3.15 |