Object Existence Hallucination Evaluation on POPE (average across three subsets)
88.93AccuracyAntidote
Evaluation results
| Method | Links | ||
|---|---|---|---|
| AntidoteBackbone=LLaVA-1.5-13B, Strategy=Antidote (Post-training)2025.04 | 88.93 | 88.99 | |
| AntidoteBackbone=LLaVA-1.5-7B, Strategy=Antidote (Post-training)2025.04 | 88.09 | 87.89 | |
| TGIFBase Model=LLaVA-1.5-7B2026.01 | 87.91 | 86.23 | |
| VolcanoBackbone=LLaVA-1.5-13B, Strategy=Post-training / Decoding2025.04 | 87.02 | 87.17 | |
| VolcanoBackbone=LLaVA-1.5-7B, Strategy=Post-training / Decoding2025.04 | 86.96 | 86.67 | |
| LLaVA-1.5-7BBackbone=LLaVA-1.5-7B2026.01 | 86.85 | 85.86 | |
| SeVaBackbone=LLaVA-1.5-7B, Strategy=Post-training / Decoding2025.04 | 86.69 | 86.66 | |
| HACLBackbone=LLaVA-1.5-7B, Strategy=Post-training / Decoding2025.04 | 86.66 | 86.2 | |
| HA-DPOBackbone=LLaVA-1.5-7B, Strategy=Post-training / Decoding2025.04 | 86.63 | 86.87 | |
| VTIBase Model=LLaVA-1.5-7B2026.01 | 86.5 | 85.9 | |
| VDDBackbone=LLaVA-1.5-7B, Strategy=Post-training / Decoding2025.04 | 86.47 | 85.13 | |
| FarSightBase Model=LLaVA-1.5-7B2026.01 | 86.1 | 80.4 | |
| LLaVA-1.5-7BBackbone=LLaVA-1.5-7B2025.04 | 85.18 | 86.07 | |
| VCDBase Model=LLaVA-1.5-7B2026.01 | 84.66 | 84.51 | |
| OPERABase Model=LLaVA-1.5-7B2026.01 | 84.2 | 85.4 | |
| LLaVA-1.5-13BBackbone=LLaVA-1.5-13B2025.04 | 84.15 | 85.67 |