DAmageNet
收藏arXiv2019-12-16 更新2024-06-21 收录
下载链接：
http://www.pami.sjtu.edu.cn/Show/56/122
下载链接
链接失效反馈官方服务：
资源简介：
DAmageNet是由上海交通大学图像处理与模式识别研究所创建的首个通用对抗数据集，包含96020个从ImageNet生成的可转移对抗样本。该数据集通过零查询黑盒攻击方式生成，平均差异仅为3.8，但错误率高达90%。DAmageNet旨在通过这些精心制作的对抗样本来评估和提升深度神经网络的鲁棒性，为研究对抗攻击和提高模型安全性提供了一个重要的基准。
DAmageNet is the first general-purpose adversarial dataset created by the Institute of Image Processing and Pattern Recognition, Shanghai Jiao Tong University. It consists of 96,020 transferable adversarial examples generated from ImageNet. This dataset is generated through zero-query black-box attack approaches, with an average difference of only 3.8 and an error rate reaching up to 90%. DAmageNet is designed to evaluate and improve the robustness of deep neural networks via these meticulously crafted adversarial samples, serving as a critical benchmark for research on adversarial attacks and enhancing model security.
提供机构：
上海交通大学图像处理与模式识别研究所
创建时间：
2019-12-16
