Object Hallucination Assessment on MSCOCO
30.2CHAIR Instance ScoremPLUG-Owl
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| mPLUG-Owl2023.10 | 30.2 | 76.8 | 98.5 | — | — | — | |
| LLaVA2023.10 | 18.8 | 62.7 | 90.7 | — | — | — | |
| MultiModal-GPT2023.10 | 18.2 | 36.2 | 45.7 | — | — | — | |
| LLaVA-1.5-7BModel Size=7B2024.05 | 14.9 | 48.8 | — | — | — | — | |
| LLaVA-1.5-13BModel Size=13B2024.05 | 14.1 | 48.3 | — | — | — | — | |
| VlfeedbackBackbone=LLaVA-1.5-7B2024.05 | 13.2 | 40.3 | — | — | — | — | |
| Human-PreferBackbone=LLaVA-1.5-7B2024.05 | 11.3 | 38.7 | — | — | — | — | |
| Multinomial SamplingBackbone=InstructBLIP, Max new tokens=642025.12 | 10.4 | 32.8 | 53.9 | — | — | — | |
| MiniGPT-42023.10 | 9.2 | 31.5 | 116.2 | — | — | — | |
| Self-rewardingBackbone=LLaVA-1.5-13B2024.05 | 8.8 | 37 | — | — | — | — | |
| MiniGPT-v2instructional_template=long2023.10 | 8.7 | 25.3 | 56.5 | — | — | — | |
| POVIDBackbone=LLaVA-1.5-7B2024.05 | 8.3 | 35.2 | — | — | — | — | |
| Multinomial SamplingBackbone=LLaVA-1.5, Max new tokens=642025.12 | 7.9 | 26.4 | 53.4 | — | — | — | |
| MiniGPT-v2instructional_template=grounded2023.10 | 7.6 | 12.5 | 18.9 | — | — | — | |
| RLHF-VBackbone=LLaVA-1.5-7B2024.05 | 7.5 | 29.7 | — | — | — | — | |
| Multinomial SamplingBackbone=LLaVA-NeXT, Max new tokens=642025.12 | 7.5 | 18 | 60.1 | — | — | — | |
| CSRBackbone=LLaVA-1.5-13B2024.05 | 7.3 | 28 | — | — | — | — | |
| Greedy DecodingBackbone=InstructBLIP, Max new tokens=642025.12 | 7.3 | 26.2 | 55.4 | — | — | — | |
| OPERABackbone=InstructBLIP, Max new tokens=642025.12 | 7.3 | 22.4 | 52.7 | — | — | — | |
| Beam SearchBackbone=InstructBLIP, Max new tokens=642025.12 | 7.1 | 25.8 | 55.3 | — | — | — | |
| HALCBackbone=InstructBLIP, Max new tokens=642025.12 | 6.8 | 23.2 | 54.7 | — | — | — | |
| Self-rewardingBackbone=LLaVA-1.5-7B2024.05 | 6.7 | 24 | — | — | — | — | |
| VCDBackbone=LLaVA-1.5, Max new tokens=642025.12 | 6.4 | 22.4 | 54.3 | — | — | — | |
| VCDBackbone=InstructBLIP, Max new tokens=642025.12 | 6.3 | 24 | 54.4 | — | — | — | |
| HDDBackbone=InstructBLIP, Max new tokens=642025.12 | 6.3 | 21.2 | 55.5 | — | — | — | |
| HALCBackbone=LLaVA-1.5, Max new tokens=642025.12 | 6.2 | 22 | 54.5 | — | — | — | |
| OPERABackbone=LLaVA-1.5, Max new tokens=642025.12 | 6.1 | 21.4 | 54.1 | — | — | — | |
| VCDBackbone=LLaVA-NeXT, Max new tokens=642025.12 | 6.1 | 15.1 | 60.2 | — | — | — | |
| CSRBackbone=LLaVA-1.5-7B2024.05 | 6 | 21 | — | — | — | — | |
| Greedy DecodingBackbone=LLaVA-1.5, Max new tokens=642025.12 | 6 | 21 | 54.7 | — | — | — | |
| Greedy DecodingBackbone=LLaVA-NeXT, Max new tokens=642025.12 | 6 | 14.8 | 60.7 | — | — | — | |
| Beam SearchBackbone=LLaVA-1.5, Max new tokens=642025.12 | 5.9 | 20.8 | 54.5 | — | — | — | |
| HALCBackbone=LLaVA-NeXT, Max new tokens=642025.12 | 5.8 | 14.9 | 60.3 | — | — | — | |
| Beam SearchBackbone=LLaVA-NeXT, Max new tokens=642025.12 | 5.7 | 14.4 | 60.4 | — | — | — | |
| OPERABackbone=LLaVA-NeXT, Max new tokens=642025.12 | 5.7 | 14.7 | 59.8 | — | — | — | |
| HDDBackbone=LLaVA-1.5, Max new tokens=642025.12 | 5.6 | 18.4 | 52.7 | — | — | — | |
| HDDBackbone=LLaVA-NeXT, Max new tokens=642025.12 | 5.5 | 13.6 | 60.2 | — | — | — | |
| MiniGPT-v2instructional_template=short2023.10 | 4.4 | 7.1 | 10.3 | — | — | — | |
| baseMLLM=LLaVA-v1.52025.09 | — | — | 99.8 | 48.8 | 13.4 | 78.6 | |
| MiddleMLLM=LLaVA-v1.52025.09 | — | — | — | 25 | 6.7 | — | |
| Ours_ES_12%MLLM=LLaVA-v1.5, Early Stopping Threshold=12%2025.09 | — | — | 54.1 | 23.5 | 6.5 | 55 | |
| PAIMLLM=LLaVA-v1.52025.09 | — | — | — | 24.8 | 6.9 | — |