Awesome
Unsupervised Image to Image Translation with Generative Adversarial Networks
<a href="http://tensorlayer.readthedocs.io">
<div align="center">
<img src="img/results.png" width="70%" height="70%"/>
</div>
</a>
Requirements
- TensorFlow 1.0.0
- TensorLayer 1.3.11
- CUDA 8
- Ubuntu
Dataset
- Before training the network, please prepare the data
- CelebA download
- Cropped SVHN download
- MNIST download, and put to
data/mnist_png
Usage
Step 1: Learning shared feature
python3 train.py --train_step="ac_gan" --retrain=1
Step 2: Learning image encoder
python3 train.py --train_step="imageEncoder" --retrain=1
Step 3: Translation
python3 translate_image.py
- Samples of all steps will be saved to data/samples/
Network
<a href="http://tensorlayer.readthedocs.io">
<div align="center">
<img src="img/network.png" width="70%" height="70%"/>
</div>
</a>
Want to use different datasets?
- in
train.py and translate_image.py modify the name of dataset flags.DEFINE_string("dataset", "celebA", "The name of dataset [celebA, obama_hillary]")
- write your own
data_loader in data_loader.py