How to run a GAN for beginners

08 Feb 2018

TLDR:

Run DCGAN on MNIST (Keras on Tensorflow)

Use the code by Rowel Atienza here:[https://github.com/roatienza/Deep-Learning-Experiments/blob/master/Experiments/Tensorflow/GAN/dcgan_mnist.py]

Blogpost here: [https://towardsdatascience.com/gan-by-example-using-keras-on-tensorflow-backend-1a6d515a60d0]

Run DCGAN on your own black and white dataset

Resize around >1500 images into 28x28 squares using the resize_centre() function in 0_split_pics_svg.py

Run DCGAN on MNIST/ Anime/ Korean Face datasets (Tensorflow)

Use the code by carpedm20 here: [https://github.com/carpedm20/DCGAN-tensorflow]

Or use the code by Lilian Weng here: [https://github.com/lilianweng/unified-gan-tensorflow]

Run Improved WGAN on MNIST (Keras on Tensorflow)

Use the code on keras-contrib here: [https://github.com/keras-team/keras-contrib/blob/master/examples/improved_wgan.py]

Run WGAN on your own B&W/RGB dataset Pre-process at least 200 images (but ideally 2000 or more) with 1_resize_to_array.py, and run on 2_WGAN_bw for your black and white dataset, or 2_WGAN_rgb for coloured dataset.