Revealing Multi-View Hallucination in Large Vision-Language Models

1. 1Construct a benchmark like MVH-Bench to evaluate multi-view hallucination
2. 2Analyze question-answer pairs to identify instances of hallucination
3. 3Develop strategies to mitigate hallucination in large vision-language models
4. 4Evaluate model performance using the benchmark to ensure accuracy and reliability

Computer vision engineers and researchers working with large vision-language models can benefit from this study to improve model performance and reduce hallucination errors

💡 Multi-view hallucination can lead to significant errors in large vision-language models, and a systematic approach is needed to address this issue

🔍 New benchmark MVH-Bench to tackle multi-view hallucination in large vision-language models