VQA Hallucination Detection on POPE Average (Random, Popular, Adversarial) 2023
89.4AccuracyCRoPS
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CRoPSModel Architecture=LLaVA-NeXT2026.01 | 89.4 | 89.4 | |
| SIDModel Architecture=LLaVA-NeXT2026.01 | 88.6 | 88.4 | |
| M3IDModel Architecture=LLaVA-NeXT2026.01 | 88 | 88.1 | |
| VCDModel Architecture=LLaVA-NeXT2026.01 | 87.7 | 88.1 | |
| OPERAModel Architecture=LLaVA-NeXT2026.01 | 87.4 | 88.2 | |
| SamplingModel Architecture=LLaVA-NeXT2026.01 | 87.3 | 87.6 | |
| ClearSightModel Architecture=LLaVA-NeXT2026.01 | 87.3 | 87.7 | |
| ICDModel Architecture=LLaVA-NeXT2026.01 | 86.6 | 87.4 | |
| CRoPSModel Architecture=Qwen2-VL2026.01 | 86.1 | 85.3 | |
| SIDModel Architecture=Qwen2-VL2026.01 | 85.6 | 83.9 | |
| OPERAModel Architecture=Qwen2-VL2026.01 | 85.5 | 84.5 | |
| ICDModel Architecture=Qwen2-VL2026.01 | 85.2 | 84.3 | |
| M3IDModel Architecture=Qwen2-VL2026.01 | 84.9 | 83.5 | |
| VCDModel Architecture=Qwen2-VL2026.01 | 84.7 | 83.2 | |
| SamplingModel Architecture=Qwen2-VL2026.01 | 84.2 | 82.8 | |
| CRoPSModel Architecture=LLaVA-1.52026.01 | 83.9 | 84.6 | |
| SIDModel Architecture=LLaVA-1.52026.01 | 82.7 | 83.3 | |
| VCDModel Architecture=LLaVA-1.52026.01 | 82.6 | 83.3 | |
| M3IDModel Architecture=LLaVA-1.52026.01 | 82.4 | 83.4 | |
| OPERAModel Architecture=LLaVA-1.52026.01 | 82.3 | 82.9 | |
| ICDModel Architecture=LLaVA-1.52026.01 | 82.2 | 82.9 | |
| ClearSightModel Architecture=LLaVA-1.52026.01 | 82 | 83 | |
| SamplingModel Architecture=LLaVA-1.52026.01 | 81.7 | 82.8 | |
| ClearSightModel Architecture=Qwen2-VL2026.01 | 73.5 | 74.5 |