Automatic Sketch Colorization with Tandem Conditional Adversarial Networks
Hui Ren, Jia Li, Nan Gao · 2018
Similar to stacked generative adversarial networks (GANs) whose each GAN undertakes one decomposition task, such as a Gaussian pyramid module or a palette module. In this paper, we use the Unet network in a tandem format to deal with the sketch-to-image problem. The first conditional GAN generates a grayscale image and the second cGAN automatically generates color comics based on the grayscale image. We use the SED model using structured random forests to obtain the original sketch. Compared with DeepColor, PaintsChainer and PaintsTransfer, it shows that the algorithm proposed in this paper can achieve higher quality automatic manga colorization which is visually more closed to the real comics.