Robustness of Visual Explanations to Common Data Augmentation