randall-lab/cifar10-c
收藏Hugging Face2025-02-23 更新2025-04-26 收录
下载链接：
https://hf-mirror.com/datasets/randall-lab/cifar10-c
下载链接
链接失效反馈官方服务：
资源简介：
CIFAR-10-C数据集是CIFAR-10的扩展版本，专门设计用于评估模型对常见腐败的鲁棒性。该数据集包含了950,000张图像，这些图像通过对原始CIFAR-10测试集中的10,000张图像应用19种不同的腐败类型和5个严重级别得到。腐败类型包括噪声、模糊、天气效果和数字扭曲等。该数据集广泛用于图像分类任务的鲁棒性基准测试。
The CIFAR-10-C dataset is an extension of the CIFAR-10 designed to evaluate model robustness to common corruptions. It consists of 950,000 images derived from the original CIFAR-10 test set (10,000 images) by applying 19 different corruption types at 5 severity levels. The corruptions include noise, blur, weather effects, and digital distortions. This dataset is widely used for benchmarking robustness in image classification tasks.
提供机构：
randall-lab
