Evaluation Metrics

ArXiv Preprint Says Prompts Drive Many LVLM Hallucinations

The authors present tools to test and reduce instruction-induced mistakes in image-text models.