arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2402.19150cs.CV
收藏

揭示排版欺骗:大型视觉-语言模型中排版脆弱性的洞察

Unveiling Typographic Deceptions: Insights of the Typographic Vulnerability in Large Vision-Language Model

The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州));University of Oxford(牛津大学);Xi’an Jiaotong University(西安交通大学);Drexel University(德雷塞尔大学);The Hong Kong University of Science and Technology(香港科技大学)

Hao Cheng, Erjia Xiao, Jindong Gu, Le Yang, Jinhao Duan, Jize Zhang, Jiahang Cao, Kaidi Xu, Renjing Xu

AI总结:

本研究验证排版攻击对大型视觉-语言模型的广泛威胁,构建迄今最大排版数据集,揭示攻击成因,并将性能损失从42.07%降至13.90%。

摘要:

大型视觉-语言模型(LVLMs)依赖视觉编码器和大型语言模型(LLMs),在视觉与语言的联合空间中展现出对各种多模态任务的卓越能力。然而,排版攻击(typographic attacks)曾被认为会破坏对比语言-图像预训练(CLIP)等视觉-语言模型(VLMs),如今也被预期为对LVLMs的一种安全威胁。首先,我们验证了排版攻击对当前知名的商业和开源LVLMs的影响,并揭示了这一威胁的广泛存在。其次,为了更好地评估这一脆弱性,我们提出了迄今为止最全面、规模最大的排版数据集(Typographic Dataset)。该数据集不仅考虑了各种多模态任务下排版攻击的评估,还评估了由多种因素生成的文本所影响的排版攻击效果。基于评估结果,我们探究了排版攻击影响VLMs和LVLMs的原因,得出了三项极具洞察力的发现。在进一步验证我们发现的合理性的过程中,我们能够将排版攻击导致的性能下降从42.07%降低至13.90%。代码和数据集可在https://github.com/ChaduCheng/TypoDeceptions获取。

英文摘要:

Large Vision-Language Models (LVLMs) rely on vision encoders and Large Language Models (LLMs) to exhibit remarkable capabilities on various multi-modal tasks in the joint space of vision and language. However, typographic attacks, which disrupt Vision-Language Models (VLMs) such as Contrastive Language-Image Pretraining (CLIP), have also been expected to be a security threat to LVLMs. Firstly, we verify typographic attacks on current well-known commercial and open-source LVLMs and uncover the widespread existence of this threat. Secondly, to better assess this vulnerability, we propose the most comprehensive and largest-scale Typographic Dataset to date. The Typographic Dataset not only considers the evaluation of typographic attacks under various multi-modal tasks but also evaluates the effects of typographic attacks, influenced by texts generated with diverse factors. Based on the evaluation results, we investigate the causes why typographic attacks impacting VLMs and LVLMs, leading to three highly insightful discoveries. During the process of further validating the rationality of our discoveries, we can reduce the performance degradation caused by typographic attacks from 42.07\% to 13.90\%. Code and Dataset are available in \href{https://github.com/ChaduCheng/TypoDeceptions}

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。

↑