Generative Hallucination Mitigation on Object HalBench (test)
49.6CHAIR_SAVISC
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| AVISCStrategy=Post-hoc hallucination mitigation, Base Model=LLaVA-1.5-7B, Venue=ACL’252026.04 | 49.6 | 12.9 | |
| VCDStrategy=Post-hoc hallucination mitigation, Base Model=LLaVA-1.5-7B, Venue=CVPR’242026.04 | 48.8 | 24.3 | |
| M3IDStrategy=Post-hoc hallucination mitigation, Base Model=LLaVA-1.5-7B, Venue=CVPR’242026.04 | 47.8 | 12.5 | |
| ICDStrategy=Post-hoc hallucination mitigation, Base Model=LLaVA-1.5-7B, Venue=ACL’242026.04 | 47.4 | 13.9 | |
| LLaVA-1.5-7B (Base)Strategy=Standard LVLM2026.04 | 46.3 | 22.6 | |
| OPERAStrategy=Post-hoc hallucination mitigation, Base Model=LLaVA-1.5-7B, Venue=CVPR’242026.04 | 45.1 | 22.3 | |
| MoDStrategy=Post-hoc hallucination mitigation, Base Model=LLaVA-1.5-7B, Venue=ACL’252026.04 | 42.6 | 12.4 | |
| EOSStrategy=Fine-tune LVLMs with external annotation data, Venue=ACL’242026.04 | 40.3 | 17.8 | |
| EAZYStrategy=Post-hoc hallucination mitigation, Base Model=LLaVA-1.5-7B, Venue=ICCV’252026.04 | 38.8 | 11.4 | |
| LLaVA-RLHFStrategy=Fine-tune LVLMs with external annotation data, Venue=ACL’242026.04 | 38.1 | 18.9 | |
| DeCoStrategy=Post-hoc hallucination mitigation, Base Model=LLaVA-1.5-7B, Venue=ICLR’252026.04 | 37.8 | 11.1 | |
| HALCStrategy=Post-hoc hallucination mitigation, Base Model=LLaVA-1.5-7B, Venue=ICML’242026.04 | 21.7 | 7.1 | |
| OctopusStrategy=Post-hoc hallucination mitigation, Base Model=LLaVA-1.5-7B, Venue=CVPR’252026.04 | 20.8 | 6.6 | |
| ONLYStrategy=Post-hoc hallucination mitigation, Base Model=LLaVA-1.5-7B, Venue=ICCV’252026.04 | 20 | 6.2 | |
| HDPOStrategy=Fine-tune LVLMs with external annotation data, Venue=ACL’252026.04 | 16.6 | 5.1 | |
| GPT-4VStrategy=Standard LVLM2026.04 | 13.6 | 7.3 | |
| RLAIF-VStrategy=Fine-tune LVLMs with external annotation data, Venue=CVPR’252026.04 | 10.5 | 5.2 | |
| PSRDStrategy=Post-hoc hallucination mitigation, Base Model=LLaVA-1.5-7B2026.04 | 10.1 | 4.1 |