发现图像令牌是幻觉主因，零训练移除特定令牌可大幅降低模型幻觉。
Hallucinatory Image Tokens: A Training-free EAZY Approach on Detecting and Mitigating Object Hallucinations in LVLMs
- 从图像令牌角度切入，识别高注意力值的幻觉源令牌。
- 仅移除1.5%的令牌即有效抑制幻觉，跨模型数据集稳定生效。
- 无需训练即可检测与缓解幻觉，适配多种视觉语言模型。
尽管大型视觉语言模型（LVLMs）潜力巨大，但仍存在对象幻觉问题，即生成内容错误引入实际不存在的对象。现有研究多聚焦于语言模型本身，本文转而关注图像输入源，发现少数高注意力图像令牌（仅占所有令牌的1.5%）是幻觉的主要驱动因素。通过移除这些幻觉令牌，可有效缓解幻觉问题，且该现象在不同模型和数据集上均成立。基于此，我们提出EAZY方法——一种无需训练的自动识别并消除幻觉图像令牌的技术。利用EAZY实现无监督幻觉检测，较先前方法提升15%；同时在保持模型性能前提下，有效缓解幻觉，并可无缝适配多种LVLM架构。
原文摘要 · Abstract (English)
Despite their remarkable potential, Large Vision-Language Models (LVLMs) still face challenges with object hallucination, a problem where their generated outputs mistakenly incorporate objects that do not actually exist. Although most works focus on addressing this issue within the language-model backbone, our work shifts the focus to the image input source, investigating how specific image tokens contribute to hallucinations. Our analysis reveals a striking finding: a small subset of image tokens with high attention scores are the primary drivers of object hallucination. By removing these hallucinatory image tokens (only 1.5% of all image tokens), the issue can be effectively mitigated. This finding holds consistently across different models and datasets. Building on this insight, we introduce EAZY, a novel, training-free method that automatically identifies and Eliminates hAllucinations by Zeroing out hallucinatorY image tokens. We utilize EAZY for unsupervised object hallucination detection, achieving 15% improvement compared to previous methods. Additionally, EAZY demonstrates remarkable effectiveness in mitigating hallucinations while preserving model utility and seamlessly adapting to various LVLM architectures.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。