Generative Adversarial Networks are today’s premier choice for Deep Learning based image generation. Being capable of generating highly realistic images, and more recently illustrating that they can be explored, GANs suggest to be powerful models in the near years.
In this Collection, we explore GANs in more detail. How do they work? Which types of GANs are out there? What problems emerge when training a GAN? Through these articles, you can learn how GANs work – and explore more contemporary approaches at the same time. Starting as an introduction to GANs, they quickly move to more advanced topics.
Let’s take a look! 🚀
Table of contents
Generative Adversarial Networks, a gentle introduction
GANs are really popular today for image generation tasks. But how do they work? Understanding the internals of a Generative Adversarial Network can be really difficult.
In this tutorial, we explore the core concepts by explaining one of the foundational GAN papers – that of Goodfellow et al. from 2014. We learn about how a GAN works conceptually (easily), yet also explore the maths. We do so intuitively – meaning that we don’t care about the formalisms, but rather about the intuitions.
By reading it, you will understand how a GAN works, what it is composed of, and how they battle together in a Minimax game.
A history of GANs
Ever since the original GAN was proposed in 2014, researchers and practitioners have improved Generative Adversarial Network architectures. Next, you’ll find our articles that cover GAN innovations, so that you can easily get up to date with the history of GANs.
- (2014) Conditional GANs, also known as cGANs.
- (2015) Deep Convolutional GANs, also known as DCGANs.
- (2018) StyleGAN.
This Person Does Not Exist – How does it work?
The website thispersondoesnotexist.com is created to showcase the inner workings of StyleGAN2, a GAN from late 2019 that is capable of generating relatively high-resolution images of people.
In this article, we take a look at this website and explore it through the lens of GANs. How do GANs work related to human face generation? What are possible use cases for GANs?
By reading it, you will understand these aspects in more detail.