Data augmentation alone can significantly improve accuracy and robustness in adversarial training. By introducing a new crop transformation called Cropshift, which enhances diversity, and a new data augmentation scheme based on Cropshift, the method achieves state-of-the-art results. The diversity and hardness of data augmentation play crucial roles in combating robust overfitting, with diversity improving both accuracy and robustness, and hardness boosting robustness at the expense of accuracy within certain limits. The new augmentation method, when combined with weight averaging, matches or even surpasses the performance of the best contemporary regularization methods for addressing robust overfitting.