Implementation of the paper Domain Translation via Latent Space Mapping in PyTorch.
Default configuration can be change in params/ for each dataset and each models.
We provide two available implementation, one based on a classical CNN, and one based on CycleGAN and Pix2pix implementation.
The 300-W facial landmark detection dataset can be downloaded here. The model expect all the images to be preprocessed using the script in datasets/Face/CreateFaceDataset.ipynb.
The Toy dataset is generated using the script in datasets/Toyset/CreateDataset.ipynb. The used set can be found un datasets/Toyset/toyset_128_swap/.
The run configuration can be found in the folder Runner. Otherwise training and testing can be done using the mains script at the root of the project.
Part of the code is based on the CycleGAN and Pix2pix repository.