DMIT

Pytorch implementation of our paper: "Multi-mapping Image-to-Image Translation via Learning Disentanglement".

Dependencies

you can install all the dependencies by

pip install -r requirements.txt

Getting Started

Datasets

Download and unzip preprocessed datasets by

Season Transfer

bash ./scripts/download_datasets.sh summer2winter_yosemite

Semantic Image Synthesis

bash ./scripts/download_datasets.sh birds

Or you can manually download them from CycleGAN and AttnGAN.

Training

Season Transfer
```
bash ./scripts/train_season_transfer.sh
```

Semantic Image Synthesis

bash ./scripts/train_semantic_image_synthesis.sh

To view training results and loss plots, run python -m visdom.server and click the URL http://localhost:8097. More intermediate results can be found in environment exp_name.

Testing

Run

bash ./scripts/test_season_transfer.sh
bash ./scripts/test_semantic_image_synthesis.sh

The testing results will be saved in checkpoints/{exp_name}/results directory.

Pretrained Models

Pretrained models can be downloaded from Google Drive or Baidu Wangpan with code 59tm.

Custom Experiment

You can implement your Dataset and SubModel to start a new experiment.

Results

Season Transfer:

Semantic Image Synthesis:

bibtex

If this work is useful for your research, please consider citing :

@inproceedings{yu2019multi,
  title={Multi-mapping Image-to-Image Translation via Learning Disentanglement},
  author={Yu, Xiaoming and Chen, Yuanqi and Liu, Shan and Li, Thomas and Li, Ge},
  booktitle={Advances in Neural Information Processing Systems},
  year={2019}
}

Acknowledgement

The code used in this research is inspired by BicycleGAN, MUNIT, DRIT, AttnGAN, and SingleGAN.

The diversity regulazation used in the current version is inspired by DSGAN and MSGAN.

Contact

Feel free to reach me if there is any questions (xiaomingyu@pku.edu.cn).