👩🏼‍🎨 Edge#175: Understanding StyleGANs

+open-sourced StyleGANS for generating photorealistic synthetic images

In this issue: 

• we explore StyleGANs;  

• we explain the original StyleGAN Paper;  

• we overview open-source StyleGANs. 

Enjoy the learning!  

💡 ML Concept of the Day: Understanding StyleGANs 

In the last few weeks, we have explored different variations of the original GAN architecture. Most of those variations focused on improvements on the discriminator models in order to train more effective generators. However, there haven’t been too many attempts to improve the original architecture of the generator model itself. This was the purpose of the Style Generative Adversarial Network or StyleGAN proposed by AI researchers from NVIDIA in 2019.  

The StyleGAN architecture proposes a series of enhancements of the architecture of the generator network to better control the image synthesis process. In other GAN variations, the generator model remains mostly a black box providing little understanding about aspects such as stochastic features or the composition of the latent space essential to the image generation process. StyleGANs borrow some ideas from