ImageNet-Patch

ImageNet-Patch数据集是由意大利卡利亚里大学、热那亚大学和Pluribus One合作创建的，旨在评估机器学习模型对抗对抗性补丁的鲁棒性。该数据集包含50,000个样本，每个样本都应用了预先优化的对抗性补丁，这些补丁经过特殊设计，能够跨不同模型进行泛化。数据集的创建过程涉及从ImageNet数据集中选取图像，并应用包含多种模型的优化过程生成的补丁。ImageNet-Patch数据集主要用于快速评估机器学习模型在面对对抗性攻击时的鲁棒性，特别是在图像分类任务中。

ImageNet-Patch dataset was collaboratively developed by the University of Cagliari (Italy), the University of Genoa, and Pluribus One, with the objective of evaluating the robustness of machine learning models against adversarial patches. This dataset consists of 50,000 samples, each equipped with a pre-optimized adversarial patch that is specially engineered to generalize across diverse machine learning models. The construction of the ImageNet-Patch dataset entails selecting images from the original ImageNet dataset and applying patches generated via an optimization process that integrates multiple models. The ImageNet-Patch dataset is primarily utilized for the rapid assessment of machine learning model robustness against adversarial attacks, especially in image classification tasks.