PyTorch implementation of TraVeLGAN: Image-to-image Translation by Transformation Vector Learning.
Add configuration file and run with :
python train.py --hparams=config_file --log=/runs/exp --device={0,1,..}
You can run the training of TraVeLGAN between two classes of CIFAR10 dataset using the cifar.json file. Download the dataset into the data folder and run the command above.
Here is an example with bird-ship (32x32) translation after 500 epochs.
Here is an example with rock beauty-toucan (128x128) translation after 10k iterations (batch size 16).
As a side note, I could not reproduce results (persistent mode collapse) without spectral normalization.