Gan

This is the final part in our series on Generative Adversarial Networks (GAN). We will write our training script and look at how to run the GAN. We will also take a look at the results we get out. Can you tell the difference between the real and generated faces?

Now that we’re able to import images into our network, we really need to build the GAN iteself. This tuorial will build the GAN class including the methods needed to create the generator and discriminator. We’ll also be looking at some of the data functions needed to make this work.

We’re ready to code! In Part 1 we looked at how GANs work and Part 2 showed how to get the data ready. In this Part, we will begin creating the functions that handle the image data including some pre-procesing and data normalisation.

This tutorial will provide the data that we will use when training our Generative Adversarial Networks. It will also take an overview on the structure of the necessary code for creating a GAN and provide some skeleton code which we can work on in the next post. If you’re not up to speed on GANs, please do read the brief introduction in Part 1 of this series on Generative Adversarial Networks.

We’ve seen that CNNs can learn the content of an image for classification purposes, but what else can they do? This tutorial will look at the Generative Adversarial Network (GAN) which is able to learn from a set of images and create an entirely new ‘fake’ image which isn’t in the training set. Why? By the end of this tutorial you’ll get know why this might be done and how to do it.