DPImageBench
收藏资源简介:
DPImageBench是由弗吉尼亚大学和微软研究院共同开发的用于差分隐私图像合成的统一基准测试平台。该平台包含了11种最新的差分隐私图像合成算法,旨在生成在保持敏感数据集特性的同时保护图像隐私的人工图像。平台提供了公平的比较环境,并采用了标准化的公共图像、模型架构和超参数设置,使用户能够方便地应用这些算法到私有的或公共的数据集上。DPImageBench对生成的合成图像质量进行了全面的评估,以便进行比较研究。
DPImageBench is a unified benchmark platform for differential privacy image synthesis, co-developed by the University of Virginia and Microsoft Research. This platform includes 11 state-of-the-art differential privacy image synthesis algorithms, aiming to generate artificial images that preserve the characteristics of sensitive datasets while protecting image privacy. It provides a fair comparison environment, adopting standardized public image datasets, model architectures and hyperparameter settings, enabling users to conveniently apply these algorithms to private or public datasets. DPImageBench conducts comprehensive evaluations on the quality of generated synthetic images to facilitate comparative research.

- 1DPImageBench: A Unified Benchmark for Differentially Private Image Synthesis弗吉尼亚大学 · 2025年



