DCGAN For CIFAR10

A simple implementation of DCGAN on CIFAR10 dataset.

Usage

1. Install all requirements by

   pip install -r requirements.txt

2. Run the following command to train the generator and discriminator. The dataset will be downloaded automatically.

   python -u main.py

3. Visualize the fake images by setting is_train = False in main.py.

4. To visualize the performance in training, run the following command.

   tensorboard --logdir=/your/summary/path

Performance

After training for 30 epochs on CIFAR10 dataset with the batch size of 128:

Reference

1. https://github.com/Ksuryateja/DCGAN-CIFAR10-pytorch
2. https://pytorch.org/tutorials/beginner/dcgan_faces_tutorial.html#training
3. https://medium.com/@stepanulyanin/dcgan-adventures-with-cifar10-905fb0a24d21
4. DCGAN paper