ImaGAN: Unsupervised Training of Conditional Joint CycleGAN for Transferring Style with Core Structures in Content Preserved

This is an official repo for "ImaGAN: Unsupervised Training of Conditional Joint CycleGAN for Transferring Style with Core Structures in Content Preserved" implemented using PyTorch. This code heavily borrows from the code for "Learning to Discover Cross-Domain Relations with Generative Adversarial Networks" by carpedm20 available here.

Download datasets (from pix2pix) with:

$ bash ./data/download_dataset.sh edges2shoes
$ bash ./data/download_dataset.sh edges2handbags

• edges2shoes: 50k training images from UT Zappos50K dataset.
• edges2handbags: 137K Amazon Handbag images from iGAN project.

Requirements

$ pip install -r requirements.txt

Start training

$ python main.py

To monitor results

$ tensorboard --logdir runs

Check points are saved in logs

tensorboard summaries are saved in runs

---

If you find our work useful please cite

@InProceedings{10.1007/978-3-030-20890-5_29,
author = {Bae, Kangmin and Ma, Minuk and Jang, Hyunjun and Ju, Minjeong and Park, Hyoungwoo and Yoo, Chang D.},
year = {2019},
month = {06},
pages = {447-462},
booktitle={Asian Conference on Computer Vision 2018},
title = {ImaGAN: Unsupervised Training of Conditional Joint CycleGAN for Transferring Style with Core Structures in Content Preserved},
isbn = {978-3-030-20889-9},
doi = {10.1007/978-3-030-20890-5_29}
}