Swapping Autoencoder for Deep Image Manipulation