Generative Hallucination on Object HalBench
61CHAIR_S ScoreLCD
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LCDBackbone=LLaVA-v1.5-7B2025.03 | 61 | 16.1 | |
| POVIDBackbone=LLaVA-v1.5-7B2025.03 | 48.1 | 24.4 | |
| ICDBackbone=LLaVA-v1.5-7B2025.03 | 47.4 | 13.9 | |
| OPERABackbone=LLaVA-v1.5-7B2025.03 | 45.1 | 22.3 | |
| EOSBackbone=LLaVA-v1.5-7B2025.03 | 40.3 | 17.8 | |
| HA-DPOBackbone=LLaVA-v1.5-7B2025.03 | 39.9 | 19.9 | |
| OPERABackbone=LLaVA-1.5-7B2026.05 | 28.3 | 12.1 | |
| LLaVA-1.5-7BBackbone=LLaVA-1.5-7B2026.05 | 27 | 10.5 | |
| EVASBackbone=LLaVA-1.5-7B2026.05 | 27 | 11.3 | |
| M3IDBackbone=LLaVA-1.5-7B2026.05 | 26.3 | 9.2 | |
| TAMEBackbone=LLaVA-1.5-7B2026.05 | 26.3 | 10.6 | |
| LLaVA-v1.5-7BBase VLM Model=LLaVA-v1.5-7B2025.05 | 26.1 | 49.7 | |
| Base (Comparable)Backbone=LLaVA-v1.5-7B2025.03 | 25 | 9.2 | |
| AVISCBackbone=LLaVA-1.5-7B2026.05 | 24.1 | 7.8 | |
| VCDBackbone=LLaVA-v1.5-7B, reproduced=true2025.03 | 23.6 | 8.4 | |
| M3IDBackbone=LLaVA-v1.5-7B, reproduced=true2025.03 | 23.2 | 7.3 | |
| DPOBase VLM Model=LLaVA-v1.5-7B, Preference Learning Method=DPO2025.05 | 23.2 | 42.3 | |
| VCDBackbone=LLaVA-1.5-7B2026.05 | 23 | 8.3 | |
| ICDBackbone=LLaVA-1.5-7B2026.05 | 22.3 | 7.4 | |
| AVISCBackbone=LLaVA-v1.5-7B, reproduced=true2025.03 | 22.1 | 7.8 | |
| LLaVA-v1.5-13BBase VLM Model=LLaVA-v1.5-13B2025.05 | 21.2 | 44.3 | |
| OctopusBackbone=LLaVA-v1.5-7B2025.03 | 20.8 | 6.6 | |
| FuzzyCDBackbone=LLaVA-1.5-7B2026.05 | 20.3 | 7.3 | |
| HA-DPOBase VLM Model=LLaVA-v1.5-7B, Preference Learning Method=HA-DPO2025.05 | 19.9 | 39.9 | |
| YARDBackbone=LLaVA-1.5-7B2026.05 | 19.7 | 6.5 | |
| DPOBase VLM Model=LLaVA-v1.5-13B, Preference Learning Method=DPO2025.05 | 19.4 | 38.3 | |
| LCDBackbone=Instruct-BLIP2025.03 | 17.4 | 10.7 | |
| OPERABackbone=Instruct-BLIP2025.03 | 16.6 | 6.8 | |
| mDPOBase VLM Model=LLaVA-v1.5-13B, Preference Learning Method=mDPO2025.05 | 16.6 | 33.3 | |
| mDPOBase VLM Model=LLaVA-v1.5-7B, Preference Learning Method=mDPO2025.05 | 16 | 30.7 | |
| ICDBackbone=Instruct-BLIP2025.03 | 15.2 | 8 | |
| LPOIBase VLM Model=LLaVA-v1.5-7B, Preference Learning Method=LPOI2025.05 | 14.6 | 24.3 | |
| GPT-4V (Base)Backbone=GPT-4V2025.03 | 13.6 | 7.3 | |
| LPOIBase VLM Model=LLaVA-v1.5-13B, Preference Learning Method=LPOI2025.05 | 11.7 | 24.3 | |
| mDPOBase VLM Model=Idefics2-8B, Preference Learning Method=mDPO2025.05 | 5.4 | 7.3 | |
| Idefics2-8BBase VLM Model=Idefics2-8B2025.05 | 4.2 | 6.3 | |
| DPOBase VLM Model=Idefics2-8B, Preference Learning Method=DPO2025.05 | 4.2 | 6 | |
| LPOIBase VLM Model=Idefics2-8B, Preference Learning Method=LPOI2025.05 | 3.6 | 5.3 | |
| M3IDBackbone=Instruct-BLIP, reproduced=true2025.03 | 0.9 | 7.6 | |
| VCDBackbone=Instruct-BLIP, reproduced=true2025.03 | 0.8 | 8.9 | |
| Base (Comparable)Backbone=Instruct-BLIP2025.03 | 0.7 | 9.1 | |
| AVISCBackbone=Instruct-BLIP, reproduced=true2025.03 | 0.7 | 8.3 | |
| OctopusBackbone=Instruct-BLIP2025.03 | 0.5 | 6.8 |