CIFAR-N

该数据集名为CIFAR-N，包括CIFAR-10N和CIFAR-100N两个部分，它们是CIFAR-10和CIFAR-100的数据集版本，由人类标注者使用多种类型的噪声标签进行了重新标注。CIFAR-10N的标签类型包括随机1、随机2、随机3、聚合和最差，而CIFAR-100N则只有一个由人类标注的标签集。这些数据集涵盖了多种类型的标签，其规模各异，针对的任务是带有噪声标签的图像分类。

This dataset is named CIFAR-N, comprising two subsets: CIFAR-10N and CIFAR-100N. Both are revised editions of the original CIFAR-10 and CIFAR-100 datasets, which were re-annotated by human annotators using multiple types of noisy labels. For CIFAR-10N, the available label types include Random 1, Random 2, Random 3, Aggregate, and Worst; in contrast, CIFAR-100N only includes a single set of human-annotated labels. These datasets feature diverse label noise patterns with varying scales, and are designed for the task of image classification with noisy labels.