Tensorflow and Keras implementation of a cycle GAN to transform ordinary photos to Monet paintings.
The same notebook can be found here.
Cycle GAN's are useful when there is an unavailability in paired training data. Training is done in a cycled fashion. This model uses the same resnet architecture which was used in the original paper. It was trained on TPU.
- Creating new pictures in the same style used by artisits that are no more.
- Transforming pictures
We can greatly improve the model by using more training data, we can experiment with different loss functions. We can use this cycleGAN with other datasets.