Unsafe2Safe：兼顾下游效用的可控图像匿名化
Unsafe2Safe: Controllable Image Anonymization for Downstream Utility
用视觉语言模型生成成对的私有/公开描述并让大语言模型产出结构化编辑指令，再以双提示词驱动扩散编辑器只重写敏感区域，这套「检测—指令—局部编辑」流程可直接迁移到数据清洗、人脸/文本脱敏和可控局部重绘任务。
原文依据 rewrites only their sensitive regions using multimodally guided diffusion editing
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论大规模图像数据集常包含可识别或敏感内容，当训练可能记忆并泄露此类信息的模型时会带来隐私风险。本文提出 Unsafe2Safe，一个全自动流水线，它检测易含隐私信息的图像，并利用多模态引导的扩散编辑仅重写其中的敏感区域。Unsafe2Safe 分两个阶段运行。阶段一使用视觉语言模型：(i) 检查图像的隐私风险；(ii) 生成成对的私有描述与公开描述，二者分别包含和省略敏感属性；(iii) 提示大语言模型以公开描述为条件，产出结构化的、身份中立的编辑指令。阶段二采用指令驱动的扩散编辑器来施加这些双重文本提示词，生成在保持全局结构与任务相关语义的同时中和私有内容的隐私安全图像。为衡量匿名化质量，我们引入一个统一的评测套件，涵盖质量、作弊、隐私和效用四个维度。在 MS-COCO、Caltech101 和 MIT Indoor67 上，Unsafe2Safe 大幅降低了人脸相似度、文本相似度和人口统计可预测性，同时保持与在原始数据上训练相当的下游模型准确率。在我们自动生成的三元组（私有描述、公开描述、编辑指令）上微调扩散编辑器，进一步提升了隐私保护与语义保真度。Unsafe2Safe 为构建大规模隐私安全数据集提供了一种可扩展且有原则的解决方案，且不牺牲视觉一致性或下游效用。
英文原文摘要 1618 字符
Large-scale image datasets frequently contain identifiable or sensitive content, raising privacy risks when training models that may memorize and leak such information. We present Unsafe2Safe, a fully automated pipeline that detects privacy-prone images and rewrites only their sensitive regions using multimodally guided diffusion editing. Unsafe2Safe operates in two stages. Stage 1 uses a vision-language model to (i) inspect images for privacy risks, (ii) generate paired private and public captions that respectively include and omit sensitive attributes, and (iii) prompt a large language model to produce structured, identity-neutral edit instructions conditioned on the public caption. Stage 2 employs instruction-driven diffusion editors to apply these dual textual prompts, producing privacy-safe images that preserve global structure and task-relevant semantics while neutralizing private content. To measure anonymization quality, we introduce a unified evaluation suite covering Quality, Cheating, Privacy, and Utility dimensions. Across MS-COCO, Caltech101 and MIT Indoor67, Unsafe2Safe reduces face similarity, text similarity, and demographic predictability by large margins, while maintaining downstream model accuracy comparable to training on raw data. Fine-tuning diffusion editors on our automatically generated triplets (private caption, public caption, edit instruction) further improves both privacy protection and semantic fidelity. Unsafe2Safe provides a scalable, principled solution for constructing large, privacy-safe datasets without sacrificing visual consistency or downstream utility.