Object Hallucination Detection on POPE averaged across MS-COCO, A-OKVQA, and GQA (Adversarial)
0.807AccuracyVAF
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VAFBackbone=LLaVA-v1.5-13B2025.03 | 0.807 | 0.817 | |
| VAFBackbone=Qwen-VL-Chat-7B2025.03 | 0.804 | 0.812 | |
| VAFBackbone=LLaVA-v1.5-7B2025.03 | 0.801 | 0.81 | |
| ICDBackbone=LLaVA-v1.5-13B2025.03 | 0.791 | 0.801 | |
| VCDBackbone=Qwen-VL-Chat-7B2025.03 | 0.788 | 0.801 | |
| ICDBackbone=LLaVA-v1.5-7B2025.03 | 0.785 | 0.799 | |
| VCDBackbone=LLaVA-v1.5-13B2025.03 | 0.782 | 0.797 | |
| VCDBackbone=LLaVA-v1.5-7B2025.03 | 0.781 | 0.796 | |
| ICDBackbone=Qwen-VL-Chat-7B2025.03 | 0.781 | 0.792 | |
| RegularBackbone=LLaVA-v1.5-13B2025.03 | 0.778 | 0.795 | |
| RegularBackbone=LLaVA-v1.5-7B2025.03 | 0.776 | 0.794 | |
| RegularBackbone=Qwen-VL-Chat-7B2025.03 | 0.772 | 0.789 |