A new method for mandala image synthesis based on WGAN-GP

Abstract

Mandala, as an ancient image art, has been found to have many unexpected applications in recent years, including use in art therapies, meditation induction and human body state assessment. Despite the omnipresent applications of convolutional neural networks in imaging Synthesis, it can be found that there is no work on Mandala Image Synthesis yet. To fill this research gap, deep learning algorithms were considered in this study. With existing research on Generative Adversarial Network (GAN), a typical GAN network called WGAN-GP was used to produce Mandala images. The generator and discriminators in the WGAN-GP network are fully trained in this study. To prove the robustness of the employed method, this study also investigates and compares the performances of WGAN-GP and other GANs. The experimental results demonstrated the effectiveness and satisfactory performance of the employed method. In brief, our completed work can effectively help provide Mandala images for further research on it.