arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2501.09833cs.CV
收藏

超出预期的擦除?概念擦除如何降低非目标概念的生成质量

Erasing More Than Intended? How Concept Erasure Degrades the Generation of Non-Target Concepts

Google Research(谷歌研究);McGill University(麦吉尔大学);Rice University(里士满大学);Google Deepmind(谷歌DeepMind)
Ibtihel Amara, Ahmed Imtiaz Humayun, Ivana Kajic, Zarana Parekh, Natalie Harris, Sarah Young, Chirag Nagpal, Najoung Kim, Junfeng He, Cristina Nader Vasconcelos… 展开作者

Ibtihel Amara, Ahmed Imtiaz Humayun, Ivana Kajic, Zarana Parekh, Natalie Harris, Sarah Young, Chirag Nagpal, Najoung Kim, Junfeng He, Cristina Nader Vasconcelos, Deepak Ramachandran, Golnoosh Farnadi, Katherine Heller, Mohammad Havaei, Negar Rostamzadeh

AI总结:

本研究提出EraseBench基准,系统分析概念擦除对非目标概念的意外影响,揭示概念纠缠导致生成质量下降的现象。

摘要:

概念擦除技术近年来因其在从文本到图像模型中移除不需要的概念方面的潜力而受到广泛关注。尽管这些方法在受控环境中常常展现出令人鼓舞的结果,但它们在现实世界应用中的鲁棒性以及是否适合部署仍不确定。在这项工作中,我们(1)识别了评估净化模型时的一个关键缺口,特别是在评估其跨不同概念维度的性能方面,以及(2)系统性地分析了文本到图像模型在擦除后的失败模式。我们聚焦于概念移除对非目标概念的意外后果,这些概念处于不同层次的相互关联关系中,包括视觉相似、二项式和语义相关的概念。为了解决这一问题,我们引入了EraseBench,一个用于评估擦除后性能的综合基准。EraseBench包含超过100个精选概念、针对性的评估提示以及一套稳健的指标,以评估擦除的有效性和副作用。我们的研究结果揭示了一种概念纠缠现象,即擦除导致对非目标概念的意外抑制,造成溢出性退化,表现为失真和生成质量的下降。

英文摘要:

Concept erasure techniques have recently gained significant attention for their potential to remove unwanted concepts from text-to-image models. While these methods often demonstrate promising results in controlled settings, their robustness in real-world applications and suitability for deployment remain uncertain. In this work, we (1) identify a critical gap in evaluating sanitized models, particularly in assessing their performance across diverse concept dimensions, and (2) systematically analyze the failure modes of text-to-image models post-erasure. We focus on the unintended consequences of concept removal on non-target concepts across different levels of interconnected relationships including visually similar, binomial, and semantically related concepts. To address this, we introduce EraseBench, a comprehensive benchmark for evaluating post-erasure performance. EraseBench includes over 100 curated concepts, targeted evaluation prompts, and a robust set of metrics to assess both effectiveness and side effects of erasure. Our findings reveal a phenomenon of concept entanglement, where erasure leads to unintended suppression of non-target concepts, causing spillover degradation that manifests as distortions and a decline in generation quality.

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。

↑