arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2408.02032cs.CVcs.AI
收藏

自省解码:缓解大型视觉-语言模型的幻觉

Self-Introspective Decoding: Alleviating Hallucinations for Large Vision-Language Models

The Hong Kong Polytechnic University(香港理工大学);Hong Kong University of Science and Technology(香港科技大学);Tencent AI Lab(腾讯AI实验室);Huazhong University of Science and Technology(华中科技大学);Tsinghua University(清华大学)

Fushuo Huo, Wenchao Xu, Zhong Zhang, Haozhao Wang, Zhicheng Chen, Peilin Zhao

AI总结:

本文提出 SID 自省解码方法,通过 CT2S 保留不重要视觉 token 以放大并减去上下文相关幻觉,在低成本、无额外知识条件下提升 LVLMs 输出忠实度。

摘要:

尽管大型视觉-语言模型(LVLMs)近年来发展迅速,但被称为“幻觉”的普遍问题已成为显著瓶颈,阻碍其在现实世界中的部署。现有方法主要从两个角度缓解该问题:一种方法利用额外知识,例如使用精心构建的数据集对 LVLMs 进行鲁棒指令微调,或采用辅助分析网络,这不可避免地带来额外成本。另一种称为对比解码的方法,通过人为扰动视觉或指令原始输入来诱发幻觉,并通过对比受扰动 LVLMs 与原始 LVLMs 的输出来缓解幻觉。然而,这些方法依赖经验性的整体输入扰动,并使推理成本翻倍。为避免这些问题,我们提出一种简单而有效的方法,名为 Self-Introspective Decoding(SID,自省解码)。我们的实证研究表明,预训练 LVLMs 能够基于先前的视觉和文本(包括指令与已生成内容)token,自省式地评估视觉 token 的重要性。我们开发了 Context and Text-aware Token Selection(CT2S,上下文与文本感知 token 选择)策略,该策略在 LVLMs 的早期层之后仅保留不重要的视觉 token,以在自回归解码过程中自适应放大由文本信息引发的幻觉。该方法确保早期层吸收的多模态知识诱发的是多模态上下文相关幻觉,而非无目的幻觉。随后,原始 token logits 减去放大后的视觉-文本关联幻觉,从而引导 LVLMs 忠实地解码。大量实验表明,SID 在各项指标上生成幻觉更少、质量更高的文本,且无需额外知识,也不会带来太多额外计算负担。

英文摘要:

While Large Vision-Language Models (LVLMs) have rapidly advanced in recent years, the prevalent issue known as the `hallucination' problem has emerged as a significant bottleneck, hindering their real-world deployments. Existing methods mitigate this issue mainly from two perspectives: One approach leverages extra knowledge like robust instruction tuning LVLMs with curated datasets or employing auxiliary analysis networks, which inevitable incur additional costs. Another approach, known as contrastive decoding, induces hallucinations by manually disturbing the vision or instruction raw inputs and mitigates them by contrasting the outputs of the disturbed and original LVLMs. However, these approaches rely on empirical holistic input disturbances and double the inference cost. To avoid these issues, we propose a simple yet effective method named Self-Introspective Decoding (SID). Our empirical investigation reveals that pretrained LVLMs can introspectively assess the importance of vision tokens based on preceding vision and text (both instruction and generated) tokens. We develop the Context and Text-aware Token Selection (CT2S) strategy, which preserves only unimportant vision tokens after early layers of LVLMs to adaptively amplify text-informed hallucination during the auto-regressive decoding. This approach ensures that multimodal knowledge absorbed in the early layers induces multimodal contextual rather than aimless hallucinations. Subsequently, the original token logits subtract the amplified vision-and-text association hallucinations, guiding LVLMs decoding faithfully. Extensive experiments illustrate SID generates less-hallucination and higher-quality texts across various metrics, without extra knowledge and much additional computation burdens.

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。

↑