peterzhousz / dcton

[CVPR 2021] Disentangled Cycle Consistency for Highly-realistic Virtual Try-On

DCTON (CVPR 2021)

Disentangled Cycle Consistency for Highly-realistic Virtual Try-On (https://arxiv.org/abs/2103.09479)

Prerequisites

• python 3.6
• pytorch 1.0.0
• torchvision 0.3.0
• cuda 10.0
• opencv

To install requirements:

conda create -n dcton python=3.6
conda activate dcton
conda install pytorch==1.0.10 torchvision==0.3.0 cuda100
pip install tensorboardX
pip install opencv-python
pip install imdb
pip install tqdm

Dataset

For data preparation, please refer to VITON.

Run the Demo

Download trained weights. Put the trained weights in the 'pretrained_model' file.

We here provide some data in 'demo_data' file for demo running.

# Demo data running
bash test.sh

License

The use of this code is restricted to non-commercial research.

Acknowledgement

Thanks for pytorch-CycleGAN-and-pix2pix for providing the useful codes.

Citation

If you think our work is useful, please feel free to cite.

@inproceedings{ge2021disentangled,
  title={Disentangled Cycle Consistency for Highly-realistic Virtual Try-On},
  author={Ge, Chongjian and Song, Yibing and Ge, Yuying and Yang, Han and Liu, Wei and Luo, Ping},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={16928--16937},
  year={2021}
}