Stable Diffusion 1.4 Unlearned Models
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/LucasBeerens/RECORD
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是基于 Stable Diffusion 1.4 文本到图像生成器的一系列未学习模型，主要关注消除特定概念，如艺术风格、物体和裸露内容。数据集包含了通过各种未学习模型方法生成的图像以及相应的攻击成功率数据。每种方法生成500张图像，使用了50个对抗性提示。该任务旨在评估概念消除方法的攻击成功率。
This dataset is a collection of unlearned models based on the Stable Diffusion 1.4 text-to-image generator, primarily focusing on erasing specific concepts such as artistic styles, objects and explicit content. The dataset contains images generated via various unlearning model methods along with corresponding attack success rate data. Each method generates 500 images using 50 adversarial prompts. The task of this dataset is to evaluate the attack success rate of concept erasure methods.
提供机构：
Authors of the paper
