Tensorflow implementation of Generative Adversarial Networks[1] for MNIST dataset
This tutorial help you walk through basic idea of GAN using MNIST data
You can download mnist data using tensorflow api.
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data/")
train_images = mnist.train.images
test_images = mnist.test.imagesA GAN network contains two nerual networks.
One is discrimator network, which will accept 28 * 28 dimensions images as input, and will ouput a probabilty indicating that this input is a real data.
The other one is generator network, which will accept 100 dimension vector as input, which is sampled from normal distribution, and will output a 28 * 28 images as fake data.
The detail algorithm is as following:
[1] Goodfellow, Ian, et al. "Generative adversarial nets." Advances in neural information processing systems. 2014.