Object Hallucination Probing on OKVQA POPE Random
86.03Accuracy (Zh)CLAIM
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| CLAIMBackbone=Qwen-VL-Chat2025.06 | 86.03 | 88.13 | 83.47 | 68.83 | — | — | 86.13 | — | 82.52 | |
| CLAIMBackbone=LLaVA-1.52025.06 | 85.97 | 86.8 | 85.63 | 83.87 | 80.63 | — | — | — | 84.58 | |
| BaselineBackbone=Qwen-VL-Chat2025.06 | 85.1 | 88.47 | 67.7 | 74.63 | — | — | 59.53 | 75.17 | 72.43 | |
| VCDBackbone=Qwen-VL-Chat2025.06 | 84.3 | — | 72.23 | 73.67 | — | — | 49.07 | 75.33 | 70.92 | |
| VCDBackbone=LLaVA-1.52025.06 | 77.93 | — | 68.17 | 72.17 | 75.67 | 70.33 | — | — | 72.85 | |
| BaselineBackbone=LLaVA-1.52025.06 | 76.13 | 91 | 63.4 | 70.3 | 75.03 | 69.23 | — | — | 70.82 |