Image changes that are done to every image both in training and in production are image preprocessing steps. This includes things like resize. Thus, preprocessing steps are applied to train, valid, and test sets.
Image changes that are done only to increase dataset size for better representations are image augmentation steps. This include things like randomly altering brightness. Thus, image augmentations are only applied to training images.
Read more on the importance of preprocessing and augmenting image data for computer vision on our blog.