From StackGAN to StackGAN++

StackGAN++ (also called StackGAN v2) is an improved version of StackGAN and was proposed by Han Zhang, Tao Xu, and Hongsheng Li, et al in their paper, StackGAN++: Realistic Image Synthesis with Stacked Generative Adversarial Networks. Compared to StackGAN, there are three main differences in the design of StackGAN++, which are as follows.

  • Multi-scale image synthesis: It uses a tree-like structure (as shown in the following diagram) in which each branch represents an individual generator network and the size of generated image increases as the tree becomes higher. The quality of the images that are generated by each branch is estimated by a different discriminator network.
  • Employment of unconditional loss: Besides ...

Get Hands-On Generative Adversarial Networks with PyTorch 1.x now with the O’Reilly learning platform.

O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.