CODE: Contrasting Self-generated Description to Combat Hallucination in Large Multi-modal Models | Read Paper on Bytez