Visual Question Answering on POPE (Adv, Pop, Ran, Avg metrics)
85.1Adv AccuracyHiDe
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| HiDeModel=Qwen2.5-VL 7B2025.09 | 85.1 | 85.4 | 86.3 | 85.6 | |
| VicropModel=Qwen2.5-VL 7B2025.09 | 84.9 | 85.1 | 85.7 | 85.2 | |
| Qwen2.5-VL 7BModel=Qwen2.5-VL 7B2025.09 | 84 | 84.4 | 85.1 | 84.5 |