Believe me, using GAN it’s one of the simplest things that i saw in data augmentations. I saw systems where data augmentation is about 80% of all system, cause in many hard scenarios you just cant get real data or in such amount that you need, so people making system that generate datasets, the simplest one is using 3d generation. For example here u can use it to generate any road you want. But there are more complex systems like emotions generation based on muscles physics, particles behavior generation and other things. So GAN is like child bicycle)