Cartoon style transfer of images with low performance deep learning

The technology of image style transfer has become increasingly important in computer vision, having achieved amazing results．However, since cartoon style is different from most art styles, there is still ample room for further improvements．This paper reviews relevant background, highlights significance of image style transfer, then focuses on cartoonGAN with limited GPU performance．Image priori information was extracted from VGG network to accelerate the learning process．CartoonGAN model was tailored to make convergence possible under low performance computing conditions due to guaranteed results．Reasonable loss function was designed to ensure overall styling effect．The network was implemented under open source deep learning framework tensorflow 2.0．Data analysis and possible improvement methods are presented．