DCGAN Tutorial - Deep Convolutional Generative Adversarial Network

A tutorial of training DCGAN with PyTorch
GANs
Published

November 16, 2020

Keywords

GANs, PyTorch, deep learning, python

What is DCGAN?

GCGAN is a Deep Convolutional Generative Adversarial Network created by Alec Radford, Luke Metz and Soumith Chintala in 2016 to train Deep Generative Adversarial Networks. In the DCGAN paper, the authors trained the network to produce fake faces of celebrities and fake bedroom images.

The architecture consists of two networks - Generator and Discriminator. Generator is the heart of GANs. It produces real looking fake images from random noise.

Screenshot 2020-11-15 at 4.18.31 PM.png

Image from the paper