[NAACL 2025 Findings] Mitigating Hallucinations in Large Vision-Language Models via Summary-Guided Decoding
