VTON-IT: Virtual Try-On using Image Translation

This paper introduces VTON-IT, a novel Virtual Try-On application that uses semantic segmentation and a generative adversarial network to produce high-resolution, natural-looking images of clothes overlaid onto segmented body regions, addressing the challenges of body size, pose, and occlusions.

Project Page | Paper | Video

Requirements

• python 3.6.13
• torch 1.1.0 (as no third party libraries are required in this codebase, other versions should work, not yet tested)
• torchvision 0.3.0
• tensorboardX
• opencv

Training Pix2pix:

 python3 train.py --label_nc 0 --no_instance --name vd2.0_2  --dataroot ./datasets/vd2.0_2 --continue_train   --gpu_ids 0,1 --batchSize 2 

Train Segmentation model

u2net_train.py

Inference

Inference.py

Reference

If you find this repo helpful, please consider citing:

@misc{adhikari2023vtonit,
      title={VTON-IT: Virtual Try-On using Image Translation}, 
      author={Santosh Adhikari and Bishnu Bhusal and Prashant Ghimire and Anil Shrestha},
      year={2023},
      eprint={2310.04558},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Acknowledgements

The authors would like to thank IKebana Solutions LLC for providing them with constant support for this research project.