أرسل هذا في رسالة قصيرة: Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decoding