🌅🏞 Edge#177: An Overview of StackGANs

+an overview of NVIDIA’s Impressive GAN Projects

In this issue: 

• we overview StackGANs; 

• we explain the original StackGAN Paper; 

• we explore NVIDIA’s Impressive GAN Projects (GauGAN, GameGAN, and GANverse3D.  

Enjoy the learning!  

💡 ML Concept of the Day: An Overview of StackGANs 

In the last issue of our series about generative adversarial neural networks (GANs), we would like to cover StackGANs. This is a variation of the standard GAN architecture proposed by Baidu in order to generate photorealistic images from textual descriptions. StackGANs are one of the first applications of GANs to multi-domain neural networks (text and images) and have become incredibly popular since the publication of the original research paper.  

StackGANs are not the first neural network model to tackle the text-to-image generation problem but were one of the first able to produce photorealistic results. Previous attempts were able to synthesize images from text inputs but lacked the details on different objects without specific spatial annotation. Generating highly detailed images in one pass from a textual description is incredibly hard. To address this challenge, StackGANs