CIFAR-10-C and ImageNet-C

The dataset used in the paper is CIFAR-10-C and ImageNet-C, which are common corruption benchmarks.

BibTex: