StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation

TensorFlow Version

2.4

Datasets

CelebA

Prepare the Dataset

coming soon

Train

If your GPU has Tensor Cores(compute capability 7.0 or higher), you can train the model as follows. Otherwise you need to set an option --nouse_mp to train.

python train.py

Test

test.ipynb to test the model.
You can test your own image changing the file path and the arguments shown in the picture below.

References

• Paper

  • StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation
    

• Repos

  • https://github.com/yunjey/stargan
    
  • https://github.com/timsainb/tensorflow2-generative-models
    
  • https://github.com/tensorflow/models/tree/master/research/deeplab
    

• Webpages

  • TFRecord and tf.train.Example
  • 今さら聞けないGAN (5) WGAN-gpの実装

• Videos

  • Inside TensorFlow: tf.data + tf.distribute