What is the motivation behind not including the original images along with their transformations when using Keras's ImageDataGenerator for augmentation? Is there a study/paper that proves this is a good idea?

More Kaoutar Ben Ahmed's questions See All
Similar questions and discussions