Uni-AdaVD：通过正交值分解实现视觉生成的通用概念擦除
Uni-AdaVD: Universal Concept Erasure for Visual Generation via Orthogonal Value Decomposition
另 1 家 · 查看机构详情
- School of Computer Science and Information Engineering, Hefei University of Technology（合肥工业大学计算机与信息工程学院）
- School of Cyber Science and Technology, University of Science and Technology of China（中国科学技术大学网络空间安全学院）
- School of Artificial Intelligence and Data Science, University of Science and Technology of China（中国科学技术大学人工智能与数据科学学院）
- Institute of Software, Chinese Academy of Sciences（中国科学院软件研究所）
机构由 AI 辅助整理，请以论文原文为准。
浏览论文内容
中文总结 AI 辅助
针对视觉生成模型吸收不良概念问题，提出Uni-AdaVD框架，将多模态注意力值空间作为干预空间，结合正交值分解与自适应擦除移位抑制目标语义方向，实验证明其在单多概念擦除且保留非目标先验方面性能强大，为视觉生成模型提供安全机制。
中文摘要 AI 辅助
视觉生成模型不可避免地会从未经筛选的预训练数据中吸收不良概念，因此概念擦除对于安全部署至关重要。然而，现有的擦除方法通常是特定于架构的，难以在保留非目标内容和生成先验的同时去除目标概念。我们提出了Uni-AdaVD，这是一种用于视觉生成的通用推理时概念擦除框架。Uni-AdaVD将多模态注意力的值空间视为统一的干预空间，并引入编码器感知的目标表示构建来跨异构文本编码器定位目标语义。它进一步将正交值分解与自适应擦除移位相结合，以抑制目标语义方向，而无需更新原始模型权重。在U-Net、DiT和自回归图像生成器以及文本到视频模型上进行的大量实验表明，在保留非目标先验的同时，在单概念和多概念擦除方面具有强大的性能。这些结果表明，Uni-AdaVD为现代视觉生成模型提供了一种高效且适应性强的安全机制。我们的代码可在此https URL上获取。
英文摘要
Visual generative models inevitably absorb undesirable concepts from uncurated pretraining data, making concept erasure essential for safe deployment. Existing erasure methods, however, are often architecture-specific and struggle to remove target concepts while preserving non-target content and generative priors. We present Uni-AdaVD, a universal inference-time concept erasure framework for visual generation. Uni-AdaVD treats the value space of multimodal attention as a unified intervention space and introduces encoder-aware target representation construction to localize target semantics across heterogeneous text encoders. It further combines orthogonal value decomposition with an adaptive erasing shift to suppress target semantic directions without updating the original model weights. Extensive experiments on U-Net-, DiT-, and autoregressive image generators, as well as text-to-video models, demonstrate strong performance on single- and multi-concept erasure while preserving non-target priors. These results suggest that Uni-AdaVD provides an efficient and adaptable safety mechanism for modern visual generative models. Our code is available at https://github.com/QifanZhou/Uni-AdaVD.
