miqbal23 / dcgan_code

Modification of original DCGAN code by Radford et.al.

Deep Convolutional GAN (DCGAN)

This is a source code of Deep Convolutional Generative Adversarial Network (DCGAN), forked from the original implementation of DCGAN as proposed by Alec Radford on his paper : http://arxiv.org/abs/1511.06434

Other implementations of DCGAN

• Lua-Torch
• Chainer
• TensorFlow

This modification is intented to work with latest version of Theano (version 1.0 and later)

Requirements :

• Python 3.5
• Theano 1.0 (recommended to use bleeding-edge version)
• CUDA drivers and CuDNN library
• Scipy
• Numpy
• Matplotlib
• Tqdm
• fuel/h5py (for CelebA faces)

Training DCGAN

1. Install all requirements
2. Modify data-dir in lib/config.py to point to directories containing data
3. For MNIST, run