Chainer GAN Trainer Demo

Implementation of GAN with Chainer's trainer using the following custom classes.

• chainer.iterators

• chainer.training.updater

• chainer.training.extensions

Custom Trainer Components

chainer.iterators.RandomNoiseIterator

Iterator that keeps producing random arrays of Gaussian or uniform distribution. It has no notion of epochs.

chainer.training.updater.GenerativeAdversarialUpdater

Updater responsible for the GAN training algorithm including forward pass, backward pass and parameter updates.

chainer.training.extensions.GeneratorSample

Extension that automatically takes random sample images and saved them to disk at any given interval.

Run

Train

python train.py --gpu 0