LA-VTON (CVPRW 2024)

Official implementation for "Artifact Does Matter! Low-artifact High-resolution Virtual Try-On via Diffusion-based Warp-and-Fuse Consistent Texture" from CVPRW 2024.

[Paper] | [Workshop Site]

Environment

We recommanded to use python version >= 3.9

conda create -n LA-VTON python=3.11
conda activate LA-VTON
conda install pytorch==2.0.0 torchvision==0.15.0 torchaudio==2.0.0 pytorch-cuda=11.7 -c pytorch -c nvidia
pip install -r requirements.txt

Download trained models

Download checkpoints from drive and put in weights/

Test full model with sample images

sh test.sh

Test each stages

CTW

python CTW/test_flow.py --dataroot={dataroot} --mode=[pair|unpair] --checkpoint_path=weights/CTW.pt

Test with sample images:

python CTW/test_flow.py --dataroot=data/sample --mode=unpair --checkpoint_path=weights/CTW.pt

CTF

python CTF/test.py --dataroot={dataroot} --mode=[pair|unpair] --warped_path={warped cloth path} --checkpoint_path=weights/CTF.pt

Test with sample images:

python CTF/test.py --dataroot=data/sample --warped_path=./result/CTW_unpair/warp/ --mode=unpair --checkpoint_path=weights/CTF.pt

Test Full dataset

data prepare

1. Download full dataset from VITON-HD_dataset

2. The inner area of clothing would affect the try-on results, so we have precessed the clothing mask for better results.

   Download and unzip new_mask.zip into {dataroot}/

3. Inference with above commands.

Citation

@InProceedings{Tseng_2024_CVPR,
    author    = {Tseng, Chiang and Chen, Chieh-Yun and Shuai, Hong-Han},
    title     = {Artifact Does Matter! Low-artifact High-resolution Virtual Try-On via Diffusion-based Warp-and-Fuse Consistent Texture},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
    month     = {June},
    year      = {2024},
    pages     = {8240-8244}
}