ImageNet-R

ImageNet-R数据集由西澳大利亚大学创建，包含15,554张对对抗性攻击具有高度抵抗力的图像，这些图像在遭受攻击后能更容易地恢复。数据集通过Flickr API收集，确保与ImageNet训练集无数据泄露。该数据集旨在为对抗性攻击和防御算法的无偏见基准测试提供支持，特别适用于评估模型在面对复杂攻击时的鲁棒性。

The ImageNet-R dataset was created by the University of Western Australia. It contains 15,554 images that exhibit strong resistance to adversarial attacks and can be more easily recovered after being targeted by such attacks. The dataset was collected via the Flickr API, ensuring no data leakage between itself and the ImageNet training set. This dataset is intended to provide an unbiased benchmark for adversarial attack and defense algorithms, and is particularly well-suited for evaluating model robustness against complex adversarial attacks.