Recommendation

• Our GAN based work for facial attribute editing - https://github.com/LynnHo/AttGAN-Tensorflow.

Tensorflow implementation of DCGAN, LSGAN, WGAN and WGAN-GP, and we use DCGAN as the network architecture in all experiments.

DCGAN: Unsupervised representation learning with deep convolutional generative adversarial networks

LSGAN: Least squares generative adversarial networks

WGAN: Wasserstein GAN

WGAN-GP: Improved Training of Wasserstein GANs

DCGAN - LSGAN

WGAN - WGAN-GP

DCGAN (left: 25 epoch, right: 50 epoch (slight mode collapse))

LSGAN (left: 25 epoch, right: 50 epoch (heavy mode collapse))

left: WGAN 50 epoch, right: WGAN-GP 50 epoch

left: WGAN 100 epoch, right: WGAN-GP 100 epoch

• tensorflow r1.2
• python 2.7

python train_mnist_dcgan.py
python train_celeba_wgan.py
python train_cartoon_wgan_gp.py
...

tensorboard --logdir=./summaries/celeba_wgan --port=6006
...

1. Mnist will be automatically downloaded
2. Celeba should be prepared by yourself in ./data/img_align_celeba/*.jpg
   • Download the dataset: https://www.dropbox.com/sh/8oqt9vytwxb3s4r/AAB06FXaQRUNtjW9ntaoPGvCa?dl=0
   • the above links might be inaccessible, the alternative is
     • img_align_celeba.zip
       • https://pan.baidu.com/s/1eSNpdRG#list/path=%2FCelebA%2FImg or
       • https://drive.google.com/drive/folders/0B7EVK8r0v71pTUZsaXdaSnZBZzg
3. The cartoon-face dataset should be prepared by yourself in ./data/faces
   • Download the dataset: https://pan.baidu.com/s/1eSifHcA, password: g5qa
   • Reference: https://zhuanlan.zhihu.com/p/24767059