-
CIFAR-10-C
CIFAR-10-C is a dataset of 60,000 32x32 color images in 10 classes, with 6,000 images per class, and 10% of the images are corrupted. -
CIFAR-100-C
The dataset used in the paper is the CIFAR-100-C dataset, which applies common corruptions such as changes in contrast or blurring to images from CIFAR-100. -
ImageNet-C
The dataset used in the paper is the ImageNet-C dataset, which is a dataset of images corrupted with various types of noise and occlusions.