Dual-Path Joint Learning Art GAN and Its Application

With the rapid development of computer vision, GAN technique based on deep learning is widely used in image generation, and it enhances the realism of synthetic images. Meanwhile GAN are also widely used in art painting image generation tasks. In order to inspire painter's creativity, improve image quality and richness, and solve the problem of few data of art painting images, this paper proposes Dual-path joint learning Art GAN(DArtGAN). Which adopts multi-scale fusion learning strategy in the generator to learn and fuse input images under different resolutions, so as to improve the learning efficiency of few data. At the same time, content image and style image are input as dual paths for joint learning to generate high-quality images with flexible styles. Experiments prove that the images generated by this method have a good improvement effect in terms of quality and style diversity, and achieve certain advantages in processing complex images like landscape paintings.