vahidfazeli / DCGAN-TensorFlow

DCGAN TensorFlow Implementation

DCGAN-TensorFlow

This repository is a Tensorflow implementation of Alec Radford's Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks, ICLR2016.

Requirements

• tensorflow 1.9.0
• python 3.5.3
• numpy 1.14.2
• pillow 5.0.0
• pickle 0.7.4
• scipy 0.19.0
• matplotlib 2.0.2

Applied GAN Structure

1. Generator

2. Discriminator

Generated Images

1. MNIST

2. CIFAR10

3. CelebA

Documentation

Download Dataset

MNIST and CIFAR10 dataset will be downloaded automatically if in a specific folder there are no dataset. Use the following command to download CelebA dataset and copy the `CelebA' dataset on the corresponding file as introduced in Directory Hierarchy information.

python download2.py celebA

Directory Hierarchy

.
│   DCGAN
│   ├── src
│   │   ├── cache.py
│   │   ├── cifar10.py
│   │   ├── dataset.py
│   │   ├── dataset_.py
│   │   ├── download.py
│   │   ├── main.py
│   │   ├── solver.py
│   │   ├── tensorflow_utils.py
│   │   ├── utils.py
│   │   └── dcgan.py
│   Data
│   ├── celebA
│   ├── cifar10
│   └── mnist

src: source codes of the DCGAN

Implementation Details

Implementation uses TensorFlow implementation to train the DCGAN. Same generator and discriminator networks are used as described in Alec Radford's paper, except that batch normalization of training mode is used in training and test mode that we found to get more stalbe results.

Training DCGAN

Use main.py to train a DCGAN network. Example usage:

python main.py --is_train=true

• gpu_index: gpu index, default: 0
• batch_size: batch size for one feed forward, default: 256
• dataset: dataset name for choice [mnist|cifar10|celebA], default: mnist
• is_train: training or inference mode, default: False
• learning_rate: initial learning rate, default: 0.0002
• beta1: momentum term of Adam, default: 0.5
• z_dim: dimension of z vector, default: 100
• iters: number of interations, default: 200000
• print_freq: print frequency for loss, default: 100
• save_freq: save frequency for model, default: 10000
• sample_freq: sample frequency for saving image, default: 500
• sample_size: sample size for check generated image quality, default: 64
• load_model: folder of save model that you wish to test, (e.g. 20180704-1736). default: None

Evaluate DCGAN

Use main.py to evaluate a DCGAN network. Example usage:

python main.py --is_train=false --load_model=folder/you/wish/to/test/e.g./20180704-1746

Please refer to the above arguments.

Citation

  @misc{chengbinjin2018dcgan,
    author = {Cheng-Bin Jin},
    title = {DCGAN-tensorflow},
    year = {2018},
    howpublished = {\url{https://github.com/ChengBinJin/DCGAN-TensorFlow}},
    note = {commit xxxxxxx}
  }

Attributions/Thanks

• This project borrowed some code from carpedm20
• Some readme formatting was borrowed from Logan Engstrom

License

Copyright (c) 2018 Cheng-Bin Jin. Contact me for commercial use (or rather any use that is not academic research) (email: [e-mail]). Free for research use, as long as proper attribution is given and this copyright notice is retained.

Related Projects

• pix2pix