Learning algorithms for vision heavily rely on data augmentation as it enhances the learning process. Having access to infinite amounts of image data is less important than having good data augmentation algorithms.