Deep Convolutional Generative Adversarial Network (DCGAN) [1] implementation on MatConvNet

*Note: Any MCN version is compatible

1. Download CelebA database at: https://drive.google.com/file/d/0B7EVK8r0v71pZjFTYXZWM3FlRnM/view?usp=sharing

   Unzip and store it to '/data' folder such that 'data/img_align_celeba/00001.bmp..'
   

2. Download Pretrained DCGAN at: https://www.dropbox.com/s/dmyqd6yfga9wsm6/net_dcgan.mat?dl=0

   Place it in '/net' folder such that 'net/net_dcgan.mat'
   

Run main_start_dcgan.m

If you use GPU, set opts.idx_gpus = 1' in line 7 and opts.install.cuda_path = 'your cuda folder' in line 11)

'main_start_dcgan.m' will perform:

1. Installation of MatConvNet (ver 1.0-beta24)
2. Generation of an image DB file fitting to MatConvNet
3. Testing a pre-trained DCGAN
4. Training a new DCGAN

*Note: All hyper parameters for training the network (e.g., learning rate) are identically set to [1]

[1] Radford, Alec, Luke Metz, and Soumith Chintala. "Unsupervised representation learning with deep convolutional generative adversarial networks." arXiv preprint arXiv:1511.06434 (2015).(https://arxiv.org/abs/1511.06434)