bhiziroglu / Generative-Adversarial-Networks

Implementation of Generative Adversarial Networks using Knet for Julia

Generative-Adversarial-Networks

Implementation of the paper Generative Adversarial Networks Using Knet Library for Julia.

Introduction

• GANs are used to generate realistic looking samples.
• MNIST model uses MLP to generate samples.
• CNN is used for other datasets.
• The model must be trained on a GPU machine.
• If the dataset does not exist in the current directory, it will be downloaded.

Usage

$ julia gan_mnist.jl

$ julia gan_faces.jl

$ julia gan_cifar.jl

NOTE: To run the code, this line should be replaced with size(w,N-1) on your current Knet installation.

Generated Samples

📝 TODO

• Output images for CIFAR-10 dataset have low resolution.

📚 Tutorial

• A tutorial for Generate Adversarial Networks can be found here.

Related Works

• Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks