AEGeAN Deeper DCGAN with AE stabilization Parallel training of generative adversarial network as an autoencoder with dedicated losses for each stage.