GAN-Collections

A implement of GAN-collection for tensorflow version. DCGAN, WGAN, WGAN_GP, LSGAN, VAE_GAN, BEGAN, EBGAN etc. More GAN models will be added to this repo.

Requirements

• tensorflow 1.3.0

• python2.7 or python3.6

• numpy 1.13.*

• scipy 0.17.0~ 0.19.1

Usages

(1)download this repo to your own directory

$ git clone https://github.com/nnUyi/GAN-Collections.git
$ cd GAN-Collections

(2)download celebA dataset and store it in the data directory(directory named data)

• celebA datasets is cropping into 64*64 size with *.png or *jpg format, this repo read image format data as input.

(3)training

#you can run any gan models using main_**gan_type** like python main_wgan.py --is_training=True
$ python main_**gan_type**.py --is_training=True 
#example
$ python main_vaegan.py --is_training=True 

(4)testing

• Anyway, in this repo, testing processing is taken during training, It samples the training results to the sample(named sample) directory, and stores session in the checkpoint(named checkpoint) directory.

Experimental Results

The experimental results is produced by BEGAN after 15 epoches training, we can clearly see that, BEGAN is stable. During training process, no collapses happend.

Contacts

Email:computerscienceyyz@163.com