Our DRO-Augment framework relies on two pillars: Wasserstein Distributionally Robust Optimization and data augmentation. W-DRO aims to improve robustness against adversarial attacks by guarding against the worst-case distribution shift. Data augmentation methods, on the other hand, enhance model robustness against common corruptions by applying transformations to input images. In our framework, we first apply a chosen data augmentation method to the training data and then minimize a Wasserstein distributionally robust loss function on the augmented samples to obtain the final predictor. The settings of the refined CIFAR-10-C and CIFAR-100-C are also included here.