Unpaired image to image traslation based on GAN networks.
Project done for the Machine Learning exam.
The project is based on CycleGAN model that allows to learn how to transform an image from one domain (style) to an another using for the training two unpaired dataset. Some additions have been made on the traditional CycleGAN.
The aim of the project was in to be able to transform an human face to a simpson version of it. Most of the time of the project was spent creating and perfecting the Simpson faces dataset and for the training tuning.
For the training phase I used the Google Cloud Service.
- Languages: Python
- Concepts: Machinge Learning, Generative Adversarial Network, Unpaired Image-to-Image Translation, CycleGan, Computer Vision
- Tools: Pytorch, Visdom, Colab
- Stack: Google Cloud