Object Hallucination Detection on POPE averaged across MS-COCO, A-OKVQA, and GQA (Random)
90.1AccuracyVAF
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VAFBackbone=LLaVA-v1.5-13B2025.03 | 90.1 | 89.9 | |
| VAFBackbone=Qwen-VL-Chat-7B2025.03 | 90 | 89.7 | |
| VAFBackbone=LLaVA-v1.5-7B2025.03 | 89.6 | 89.3 | |
| VCDBackbone=Qwen-VL-Chat-7B2025.03 | 89.1 | 88.4 | |
| VCDBackbone=LLaVA-v1.5-13B2025.03 | 88.9 | 87.8 | |
| ICDBackbone=Qwen-VL-Chat-7B2025.03 | 88.9 | 88.1 | |
| VCDBackbone=LLaVA-v1.5-7B2025.03 | 88.4 | 87.7 | |
| RegularBackbone=Qwen-VL-Chat-7B2025.03 | 88.2 | 87.9 | |
| ICDBackbone=LLaVA-v1.5-7B2025.03 | 88.1 | 87.6 | |
| ICDBackbone=LLaVA-v1.5-13B2025.03 | 88.1 | 87.6 | |
| RegularBackbone=LLaVA-v1.5-7B2025.03 | 87.8 | 87.5 | |
| RegularBackbone=LLaVA-v1.5-13B2025.03 | 87.6 | 87.4 |