arXivDaily arXiv每日学术速递 周一至周五更新
​通知:联邦节假日调休:哥伦布日 / 原住民日(Columbus Day / Indigenous Peoples' Day),今日(13日)无更新。
arXiv 2507.04599cs.CV

QR-LoRA:通过 QR 分解实现高效且解耦的微调以用于定制化生成

QR-LoRA: Efficient and Disentangled Fine-tuning via QR Decomposition for Customized Generation

Harbin Institute of Technology(哈尔滨工业大学);Li Auto(利汽车);University of Science and Technology of China(中国科学技术大学)

Jiahui Yang, Yongjia Ma, Donglin Di, Hao Li, Wei Chen, Yan Xie, Jianxun Cui, Xun Yang, Wangmeng Zuo

AI总结:

针对多 LoRA 融合时内容与风格特征纠缠的问题,本文提出 QR-LoRA,利用 QR 分解固定 Q、R 并仅训练 $ΔR$,以一半可训练参数实现无交叉污染的高效解耦定制生成。

摘要:

现有的文本到图像模型通常依赖低秩适配(LoRA)等参数微调技术来定制视觉属性。然而,在组合多个 LoRA 模型以完成内容-风格融合任务时,对权重矩阵的非结构化修改往往会导致内容属性与风格属性之间出现不期望的特征纠缠。我们提出 QR-LoRA,这是一种利用 QR 分解进行结构化参数更新的新型微调框架,可有效分离视觉属性。我们的关键洞见是,正交 Q 矩阵天然能够最小化不同视觉特征之间的干扰,而上三角 R 矩阵则能高效编码特定于属性的变换。我们的方法同时固定 Q 矩阵和 R 矩阵,仅训练一个额外的任务特定 $ΔR$ 矩阵。这种结构化设计将可训练参数减少到传统 LoRA 方法的一半,并且由于 $ΔR$ 矩阵之间具有很强的解耦特性,支持有效合并多个适配而不会产生交叉污染。实验表明,QR-LoRA 在内容-风格融合任务中实现了更优的解耦效果,为生成模型中参数高效、解耦的微调建立了新范式。项目页面位于:https://luna-ai-lab.github.io/QR-LoRA/。

英文摘要:

Existing text-to-image models often rely on parameter fine-tuning techniques such as Low-Rank Adaptation (LoRA) to customize visual attributes. However, when combining multiple LoRA models for content-style fusion tasks, unstructured modifications of weight matrices often lead to undesired feature entanglement between content and style attributes. We propose QR-LoRA, a novel fine-tuning framework leveraging QR decomposition for structured parameter updates that effectively separate visual attributes. Our key insight is that the orthogonal Q matrix naturally minimizes interference between different visual features, while the upper triangular R matrix efficiently encodes attribute-specific transformations. Our approach fixes both Q and R matrices while only training an additional task-specific $ΔR$ matrix. This structured design reduces trainable parameters to half of conventional LoRA methods and supports effective merging of multiple adaptations without cross-contamination due to the strong disentanglement properties between $ΔR$ matrices. Experiments demonstrate that QR-LoRA achieves superior disentanglement in content-style fusion tasks, establishing a new paradigm for parameter-efficient, disentangled fine-tuning in generative models. The project page is available at: https://luna-ai-lab.github.io/QR-LoRA/.

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。

↑