Multi-Object Hallucination in Vision-Language Models