Kestrel: 为降低LVLM幻觉而引入自反思
Kestrel: Grounding Self-Refinement for LVLM Hallucination Mitigation
AI总结：
Kestrel提出一种无需训练的框架，通过显式视觉 grounding 与证据验证自反思机制减少LVLM幻觉，实验显示在POPE和MME-Hallucination基准上性能提升，同时提供透明的验证轨迹。
摘要：
大型视觉-语言模型（LVLMs）虽然能力增强，但在多模态任务中仍易产生幻觉，限制了其应用。随着训练这些模型以避免幻觉变得对大模型来说成本过高，训练自由方法提供了一种经济且灵活的解决方案，但现有基于解码或工具使用的方法往往收益有限或可解释性差。我们提出Kestrel，一种无需训练的LVLM幻觉缓解框架，结合显式视觉 grounding agent与证据验证的自反思机制。具体而言，Kestrel首先收集显式视觉证据并将工具输出转换为可重用的结构化文本证据。其次，通过LVLM法官验证这些证据，然后基于验证过的证据迭代自反思答案以降低过纠正的风险。大量实验表明，Kestrel在幻觉基准上优于强基线（例如，在POPE上平均提升+3.31%，在MME-Hallucination上提升+28.34，使用Qwen3-VL），同时提供透明的验证轨迹用于幻觉诊断和分析——例如，集成的自反思模块和grounding agent在POPE上平均各提升+2.0%。
英文摘要：
Large vision-language models (LVLMs) have become increasingly strong but remain prone to hallucinations in multimodal tasks, which significantly narrows their deployment. As training these LVLMs to avoid hallucinations becomes prohibitively expensive for larger models, training-free methods offer a cheap and flexible solution to this problem, yet existing approaches based on decoding or tool use often bring limited gains and/or weak interpretability. We propose Kestrel, a training-free framework for LVLM hallucination mitigation that combines an explicit visual-grounding agent with evidence-verified self-refinement mechanism. In detail, Kestrel first collects explicit visual evidence and converts tool outputs into reusable and structured textual evidence. Second, to take full advantage of these evidence, Kestrel verifies them via an LVLM judge for evidence checking, then iteratively self-refine answers based on verified evidence to reduce the risk of over-correction. Extensive experiments show that Kestrel improves performance over strong baselines across hallucination benchmarks (e.g., average +3.31% on POPE and +28.34 on MME-Hallucination with Qwen3-VL), while providing transparent verification traces for hallucination diagnosis and analysis -- e.g., both the integrated self-refinement module and grounding agent contributing an average +2.0% gain on POPE.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
