In this chapter, we will push the label-based image generation to the next level: we will use pixel-wise labeling to perform image-to-image translation and transfer image styles.
You will learn how to use pixel-wise label information to perform image-to-image translation with pix2pix and translate high-resolution images with pix2pixHD. Following this, you will learn how to perform style transfer between unpaired image collections with CycleGAN.
By the end of this chapter, combined with the knowledge from the previous chapter, you will have grasped the core methodology of using image-wise and pixel-wise label information to improve the quality, or manipulate the attributes, of generated images. You will also know how to flexibly design model architectures to accomplish your goals, including generating larger images or transferring...
