Omni-Woodpecker：通过证据-先验重平衡纠正全模态大模型幻觉
原标题：Paper page - Omni-Woodpecker: Hallucination Correction via Evidence-Prior Rebalancing for Omni LLMs
AI 摘要
研究者提出 Omni-Woodpecker（OWP），一个无需训练的框架，用于减少全模态大模型（OLLMs）中的跨模态幻觉。该方法先识别跨模态冲突，再通过正交分解将隐藏状态中的 grounding 证据与误导性先验分离，并在解码前自适应重平衡融合表示。以 Qwen2.5-Omni-7B 为骨干，OWP 在 CMM 和 AVHBench 上分别提升 12.9 和 8.6 分，在四个通用推理基准上提升 0.67-2.25 分，并可泛化到 VideoLLaMA2-AV。
正文节选
Omni-Woodpecker: Hallucination Correction via Evidence-Prior Rebalancing for Omni LLMs Abstract Omni-Woodpecker is a training-free framework that reduces cross-modal hallucinations in omni LLMs by identifying modality conflicts and rebalancing hidden representations to suppress misleading priors. Recent Omni LLMs (OLLMs) have demonstrated remarkable capabilities in joint audio-visual reasoning, yet cross-modal hallucinations remain a major challenge. Existing mitigation methods mainly operate at