Generative Hallucination Mitigation on MMHal-Bench
3.49Overall ScoreGPT-4V
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GPT-4VParadigm=Standard LVLM2026.04 | 3.49 | 28 | |
| PSRDParadigm=Post-hoc hallucination mitigation, Base Model=LLaVA-1.5-7B2026.04 | 2.92 | 49 | |
| HSA-DPOParadigm=Fine-tune LVLMs with external annotation data2026.04 | 2.61 | 48 | |
| OctopusParadigm=Post-hoc hallucination mitigation, Base Model=LLaVA-1.5-7B2026.04 | 2.61 | 50 | |
| LLaVA-DPOParadigm=Fine-tune LVLMs with external annotation data2026.04 | 2.58 | 50 | |
| SENAParadigm=Fine-tune LVLMs with self-improvement2026.04 | 2.33 | 52 | |
| AVISCParadigm=Post-hoc hallucination mitigation, Base Model=LLaVA-1.5-7B2026.04 | 2.19 | 59 | |
| OPERAParadigm=Post-hoc hallucination mitigation, Base Model=LLaVA-1.5-7B2026.04 | 2.15 | 54 | |
| M3IDParadigm=Post-hoc hallucination mitigation, Base Model=LLaVA-1.5-7B2026.04 | 2.14 | 61 | |
| STICParadigm=Fine-tune LVLMs with self-improvement2026.04 | 2.07 | 56 | |
| EOSParadigm=Fine-tune LVLMs with external annotation data2026.04 | 2.03 | 59 | |
| VCDParadigm=Post-hoc hallucination mitigation, Base Model=LLaVA-1.5-7B2026.04 | 1.96 | 64 | |
| LLaVA-1.5-7B (Base)Paradigm=Standard LVLM2026.04 | 1.55 | 76 |