Object Hallucination Probing on COCO POPE (Adversarial)
83.27Score (Zh)CLAIM
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| CLAIMBackbone=LLaVA-1.52025.06 | 83.27 | 77.67 | — | 79.43 | 79.67 | 74.57 | — | — | 78.92 | |
| BaselineBackbone=Qwen-VL-Chat2025.06 | 80.87 | 83.9 | 67.53 | 72.8 | — | — | 63.1 | 75.9 | 72.04 | |
| CLAIMBackbone=Qwen-VL-Chat2025.06 | 80.7 | 82.33 | 77.53 | 69.7 | — | — | 81.5 | — | 78.35 | |
| VCDBackbone=Qwen-VL-Chat2025.06 | 79.4 | — | 71.97 | 70.83 | — | — | 54.73 | 75.03 | 70.39 | |
| VCDBackbone=LLaVA-1.52025.06 | 74.27 | — | 67.3 | 66.47 | 73.63 | 66.33 | — | — | 69.6 | |
| BaselineBackbone=LLaVA-1.52025.06 | 73.4 | 85.2 | 62.87 | 66.07 | 73.7 | 65.93 | — | — | 68.39 |