Training WGAN-GP to Generate Fake People portrait images

Author(s): Rokas Liuberskis

Originally published on Towards AI.

We’ll cover more advanced Generative Adversarial Network techniques WGAN-GP. We’ll train it on a popular celebA human portrait dataset.

In the previous post, we covered simple DCGAN, and I gave you a simple example with the MNIST dataset. But it was a simple dataset, and we could not see the disadvantages of this architecture. The main disadvantage is its training instability; over time, it gives us worse and worse generated images.

So, to address this issue, we will implement a WGAN-GP (Wasserstein GAN with Gradient Penalty) architecture. We will train the WGAN-GP model to generate people’s profile 64×64 images by training it on the popular CelebA dataset.

The Wasserstein GAN, or WGAN, was a breakthrough in Generative Adversarial Network (GAN) training… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI