Object Hallucination Probing on OKVQA POPE Adversarial
79.97POPE Score (Zh)Baseline
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| BaselineBackbone=Qwen-VL-Chat2025.06 | 79.97 | 82.4 | 66.8 | 72.5 | — | — | 60.17 | 72.3 | 70.35 | |
| VCDBackbone=Qwen-VL-Chat2025.06 | 78.77 | — | 69 | 69.63 | — | — | 50.8 | 71.9 | 68.02 | |
| CLAIMBackbone=Qwen-VL-Chat2025.06 | 77.2 | 81.1 | 78.73 | 67.73 | — | — | 78.87 | — | 76.73 | |
| CLAIMBackbone=LLaVA-1.52025.06 | 75.4 | 77.7 | 76.17 | 75.67 | 71.97 | — | — | — | 75.38 | |
| VCDBackbone=LLaVA-1.52025.06 | 68.97 | — | 65.17 | 63.9 | 68.67 | 64.07 | — | — | 66.16 | |
| BaselineBackbone=LLaVA-1.52025.06 | 64.4 | 79.57 | 62.37 | 63.93 | 67.97 | 63.03 | — | — | 64.34 |