mathaiml5/adversarial-vision-transformers

该数据集包含两个配置：detailed和summary。detailed配置提供了详细的图像攻击和防御信息，包括原始图像索引、模型名称、攻击类型、防御类型、原始标签名称、预测标签名称、真实类别的概率、预测类别的概率、攻击是否成功、防御是否成功、输入图像、输出图像以及预测是否正确等特征。summary配置则提供了模型名称、攻击类型、防御类型、准确率、攻击成功率（ASR）和样本数量等汇总信息。数据集主要用于研究图像攻击和防御的效果。

The dataset contains two configurations: detailed and summary. The detailed configuration provides detailed information on image attacks and defenses, including original image index, model name, attack type, defense type, original label name, predicted label name, class probability of true class, class probability of predicted class, successful attack, successful defense, input image, output image, and whether the prediction is correct. The summary configuration provides aggregated information such as model name, attack type, defense type, accuracy, attack success rate (ASR), and number of samples. The dataset is primarily used for studying the effectiveness of image attacks and defenses.