MIBench
收藏资源简介:
MIBench是由哈尔滨工业大学(深圳)和清华大学共同创建的首个模型反转攻击和防御的综合基准。该数据集包含4个广泛认可的人脸数据集,包括Flickr-Faces-HQ (FFHQ)、MetFaces、FaceScrub和CelebFaces Attributes (CelebA)。这些数据集主要用于模型反转攻击的实验,支持低分辨率和高分辨率的图像处理。数据集的创建旨在为研究人员提供一个可扩展和可复现的工具箱,以标准化和公平地评估模型反转攻击和防御方法,从而推动该领域的进一步发展。
MIBench is the first comprehensive benchmark for model inversion attacks and defenses, co-developed by Harbin Institute of Technology (Shenzhen) and Tsinghua University. This dataset includes four widely recognized face datasets, namely Flickr-Faces-HQ (FFHQ), MetFaces, FaceScrub, and CelebFaces Attributes (CelebA). These datasets are primarily utilized for model inversion attack experiments, and support both low-resolution and high-resolution image processing tasks. The creation of this dataset aims to provide researchers with a scalable and reproducible toolkit to standardize and fairly evaluate model inversion attack and defense methods, thus promoting further development of this research field.




