pbcquoc / cifar_dcgan

A simple implementation of DCGAN for CIFAR 10

DCGAN for CIFAR10

A clean implementation of DCGAN for CIFAR 10 from Generative Adversarial Networks in tensorflow 1.1

PREREQUISITES

• Python 2.7
• Tensorflow 1.1
• numpy

TRAINING

You have to download cifar10 into data folder

./download.sh

To train model

python cifar_gan.py

You can find generated images at out folder

RESULTS

at 5 iters

at 85 iters

at 185 iters

at 285 iters

at 385 iters

You can see it begin with noise image at first iteration and will generate image more clearly but in the end, the model collapse happen, images are destroyed