This is the Tensorflow impementation of LS-GAN penalized with the gradient with respect to the input. IT has been implemented based on the DC-GAN implementation by Taehoon Kim.
Please cite the following paper when referring to the following algorithms:
Guo-Jn Qi. Loss-Sensitive Generative Adversarial Networks on Lipschitz Densities. arXiv:1701.06264 [pdf]
Code: https://github.com/guojunq/lsgan
Paper: https://arxiv.org/pdf/1701.06264
The implementation follows the idea of gradient penalty proposed originally in the first version of LS-GAN, Chapter 5 [pdf], quoted here "Alternatively, one may consider to directly minimize the gradient norm ||∇xLθ(x)|| as a regularizer for the LS-GAN. In this paper, we adopt weight decay for its simplicity and find it works well with the LS-GAN model in experiments."
The code has been written in Tensorflow. Following is the list of prerequisites
-Python 2.7 or Python 3.3+
-Tensorflow 0.12.1
-SciPy
-pillow Usage
Change the directory to LS-GAN-GP directory. You can download the CelebA data-set using command:
$ python download.py celebA
To train the network for CelebA dataset use the command
$ python main.py --dataset celebA --input_height=108 --train --crop
During the training process the snapshot of the models are saved in folder called checkpoint and generated samples in folder called samples
