Hallucination assessment on Object-HalBench
2.6Mention Hallucination RateOmniLMM + RLAIF-V BoN
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| OmniLMM + RLAIF-V BoNSize=12B, Feedback=self2024.05 | 2.6 | 4.5 | — | — | |
| OmniLMM + RLAIF-VSize=12B, Feedback=self2024.05 | 2.9 | 4.5 | — | — | |
| LLaVA 1.5 + RLAIF-V BoNSize=7B, Feedback=LLaVA-NeXT2024.05 | 3.8 | 6.8 | — | — | |
| MiniCPM-Llama3-V 2.5Size=8.5B2024.08 | 5 | — | — | 10.3 | |
| LLaVA 1.5 + RLAIF-VSize=7B, Feedback=LLaVA-NeXT2024.05 | 5.2 | 10.5 | — | — | |
| LLaVA-NeXTSize=34B, Feedback=X2024.05 | 6.4 | 12.6 | — | — | |
| DeepSeek-VL-7BSize=7.3B2024.08 | 6.5 | — | — | 11.4 | |
| GPT-4V (2023.11.06)2024.08 | 7.3 | — | — | 13.6 | |
| GPT-4VSize=Unknown, Feedback=Unknown2024.05 | 7.3 | 13.6 | — | — | |
| GPT-4V2023.12 | 7.3 | — | — | 13.6 | |
| RLHF-VSize=13B, Feedback=Human2024.05 | 7.5 | 12.2 | — | — | |
| RLHF-VBackbone=Vicuna v1.0 13B, Visual Module=BEiT-3, Resolution=4482023.12 | 7.5 | — | — | 12.2 | |
| MiniCPM-V 2.0Size=2.8B2024.08 | 7.8 | — | — | 14.5 | |
| MiniGeminiSize=34B, Feedback=X2024.05 | 8 | 14.5 | — | — | |
| DeepSeek-VL-1.3BSize=1.7B2024.08 | 9.6 | — | — | 16.7 | |
| OmniLMMSize=12B, Feedback=X2024.05 | 10.9 | 19.4 | — | — | |
| MiniCPM-V 1.0Size=2.8B2024.08 | 11.5 | — | — | 21.6 | |
| Yi-VL-6BSize=6.7B2024.08 | 11.7 | — | — | 19.4 | |
| CogVLM-ChatSize=17B2024.08 | 12.6 | — | — | 26.4 | |
| SilkieSize=10B, Feedback=GPT-4V2024.05 | 13.4 | 27.1 | — | — | |
| Yi-VL-34BSize=34B2024.08 | 14 | — | — | 20.7 | |
| InstructBLIP2023.12 | 14.3 | — | — | 25.9 | |
| Less-is-moreSize=7B, Feedback=X2024.05 | 17.8 | 40.3 | — | — | |
| LLaVA-RLHFSize=13B, Feedback=Human2024.05 | 18.9 | 38.1 | — | — | |
| LLaVA-RLHF2023.12 | 18.9 | — | — | 38.1 | |
| HA-DPOSize=7B, Feedback=Rule2024.05 | 19.9 | 39.9 | — | — | |
| Qwen-VL-ChatSize=9.6B2024.08 | 20 | — | — | 43.8 | |
| Qwen-VL-Chat2023.12 | 20 | — | — | 43.8 | |
| AMP-MEGSize=13B, Feedback=Rule2024.05 | 20.6 | 31.7 | — | — | |
| Qwen-VL-ChatSize=10B, Feedback=X2024.05 | 20.7 | 40.4 | — | — | |
| OPERASize=7B, Feedback=X2024.05 | 22.3 | 45.1 | — | — | |
| LRV2023.12 | 22.3 | — | — | 32.3 | |
| LLaVA 1.52023.12 | 22.6 | — | — | 46.3 | |
| CCA-LLaVASize=7B, Feedback=X2024.05 | 23.8 | 46.7 | — | — | |
| VCDSize=7B, Feedback=X2024.05 | 24.3 | 48.8 | — | — | |
| POVIDSize=7B, Feedback=Rule2024.05 | 24.4 | 48.1 | — | — | |
| Muffin2023.12 | 24.5 | — | — | 50.5 | |
| LLaVA 1.5Size=7B, Feedback=X2024.05 | 27.8 | 54.5 | — | — | |
| LLaVA2023.12 | 29.5 | — | — | 63 | |
| LLaVAParameters=13B2024.03 | — | 63 | 29.5 | — | |
| LLaVA1.5Parameters=7B2024.03 | — | 45.9 | 23.7 | — | |
| LLaVA1.5Parameters=13B2024.03 | — | 45.2 | 21.8 | — | |
| LLaVA1.5-BPOParameters=7B2024.03 | — | 31.9 | 15.1 | — | |
| LLaVA1.5-BPOParameters=13B2024.03 | — | 27.3 | 12.9 | — |