arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2510.00293cs.CVcs.CRcs.LG
收藏

MOLM:LoRA标记的混合

MOLM: Mixture of LoRA Markers

Mohamed bin Zayed University of Artificial Intelligence (MBZUAI)(穆罕默德·本·扎耶德人工智能大学);Michigan State University (MSU)(密歇根州立大学)

Samar Fares, Nurbek Tastan, Noor Hussein, Karthik Nandakumar

AI总结:

本文提出MOLM,一种基于路由的LoRA标记方法,通过轻量级LoRA适配器在残差和注意力块中实现参数扰动,提升水印的鲁棒性和隐蔽性。

摘要:

生成模型可以大规模生成逼真图像,这引发了对检测合成图像并追溯其来源的迫切需求。尽管水印技术成为可能的解决方案,但现有方法仍易受真实失真影响,易被自适应移除,并且当底层水印密钥变化时更新成本高。我们提出一个通用的水印框架,将编码问题建模为生成模型参数的关键依赖扰动。在此框架下,我们引入Mixture of LoRA Markers (MOLM),一种基于路由的实例化方法,其中二进制密钥在残差和注意力块中激活轻量级LoRA适配器。此设计避免了密钥特定的重新训练,实现了所需的属性,如不可察觉性、保真度、可验证性和鲁棒性。在Stable Diffusion和FLUX上的实验表明,MOLM在保持图像质量的同时,能够对失真、压缩和再生、平均攻击和提取器的黑盒对抗攻击实现稳健的密钥恢复。

英文摘要:

Generative models can generate photorealistic images at scale. This raises urgent concerns about the ability to detect synthetically generated images and attribute these images to specific sources. While watermarking has emerged as a possible solution, existing methods remain fragile to realistic distortions, susceptible to adaptive removal, and expensive to update when the underlying watermarking key changes. We propose a general watermarking framework that formulates the encoding problem as key-dependent perturbation of the parameters of a generative model. Within this framework, we introduce Mixture of LoRA Markers (MOLM), a routing-based instantiation in which binary keys activate lightweight LoRA adapters inside residual and attention blocks. This design avoids key-specific re-training and achieves the desired properties such as imperceptibility, fidelity, verifiability, and robustness. Experiments on Stable Diffusion and FLUX show that MOLM preserves image quality while achieving robust key recovery against distortions, compression and regeneration, averaging attacks, and black-box adversarial attacks on the extractor.

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。

↑