Code for virtual try-on with high-fidelity details. The code was developed and tested with Pytorch0.4.1.

• Clone this repo

git clone https://github.com/AIprogrammer/Detailed-virtual-try-on.git. 
cd Detailed-virtual-try-on

• Download our pretrained models from Google Drive, and put them in "./pretrained_checkpoint".

• We provide a demo model, as well as some samples in "./dataset/images". Triplets including source image, target pose, target cloth is provided in the "./demo/demo.txt".
• Quick testing and checking results in "./demo/forward/0.jpg" by running

sh demo.sh

• Download the MPV dataset from Image-based Multi-pose Virtual Try On and put the dataset under "./dataset/images/".
• Select postive perspective images, create dataset split file 'data_pair.txt', and put it under "./dataset/".

• Pose keypoints. Use the Openpose, and put the keypoints file in "./dataset/pose_coco".
• Semantic parsing. Use the CIHP_PGN, and put the parsing results in "./dataset/parse_cihp".
• Cloth mask. Use the "GrabCut" method for the cloth mask, and put the mask in "./dataset/cloth_mask".

• Download the VGG19 pretrained checkpoint

cd vgg_model/
wget https://download.pytorch.org/models/vgg19-dcbb9e9d.pth

• Set different configuration based on the "config.py". Then run

sh train.sh

If you find this code helpful, please cite our paper:

@inproceedings{detail2019,
  title={Down to the Last Detail: Virtual Try-on with Detail Carving},
  author={Wang, Jiahang and Zhang, Wei and Weizhong, Liu and Mei, Tao},
  booktitle = {arXiv:1912.06324},
  year={2019}
}