FIGURE 1.

(a) The architecture of the style‐encoding generative adversarial network (GAN). The generator learns to generate an image by inputting a source image and a style code. The style code s is generated by a mapping network from sampling a given latent vector, or encoded by a style encoder. The quality of the generated image is controlled by the discriminator which learns a binary classification determining whether an image is a real image or a fake image. (b) The detailed architecture of the generator in the network. In each of the blocks in the process, the three numbers represent the number of input channels, number of output channels, and the image size. The style code is injected into different layers of the generators to control various levels of detail and style features in the synthesized image. An instance normalization normalizes a mini‐batch of data across each channel for each observation independently.