Does Generative Adversarial Network (GAN) help in SRAF image generation?

We propose a (Deep Neural Network) DNN based method that applying Generative Adversarial Network (GAN) to the SRAF image generation. We found that though the model with GAN loss may increase distortion, it can generate high fidelity images efficiently and mitigate the overfitting issues that widely exists in all kinds of deep learning methods. We also discuss the necessity of tuning the model structure when GAN loss is adopted.