CIFAR-10N
收藏arXiv2025-09-30 收录
官方服务：
资源简介：
该数据集名为CIFAR-10N，是CIFAR-10数据集的一个变体，其中包含了标签中不同级别的噪声。在实验中，我们考虑了聚合噪声、随机1噪声和最坏情况噪声这几种情况。每个类别下有5000个训练样本。该数据集的任务包括图像分类和标签噪声检测。
This dataset, named CIFAR-10N, is a variant of the CIFAR-10 dataset that incorporates label noise at varying levels. Three types of label noise are considered in our experimental setup: aggregated noise, random 1 noise, and worst-case noise. Each class contains 5000 training samples. The tasks supported by this dataset include image classification and label noise detection.
