This repository is the official implementation of What to Preserve and What to Transfer: Faithful, Identity-Preserving Diffusion-based Hairstyle Transfer. Note that this repository only includes the inference codes.

Install PyTorch and other dependencies:

conda create -y -n hairfusion python=3.8
conda activate hairfusion
pip install torch==2.0.0+cu117 torchvision==0.15.1+cu117 --index-url https://download.pytorch.org/whl/cu117
pip install einops
pip install pytorch-lightning==1.5.0
pip install opencv-python==4.7.0.72
pip install matplotlib
pip install omegaconf
pip install albumentations
pip install transformers==4.33.2
pip install xformers==0.0.19
pip install triton==2.0.0
pip install open-clip-torch==2.19.0
pip install clean-fid==0.1.35
pip install diffusers==0.20.2
pip install scipy==1.10.1
conda install -c anaconda ipython -y

sample data in ./data/test

Preprocess a given image and save the outputs in ./data/${dir_name}

• Download face_segment16.pth (50.8MB) and shape_predictor_68_face_landmarks.dat (95.1MB) into ./models
• run

  CUDA_VISIBLE_DEVICES=0 python preprocess.py --img_path ${image_path} --save_dir_name ${dir_name} --crop_scale ${crop_scale}

Extract DensePose of the images in ./data/${dir_name}/images and save them in ./data/${dir_name}/images-densepose

• Installation

  git clone https://github.com/facebookresearch/detectron2.git
  pip install opencv-python torchgeometry Pillow tqdm tensorboardX scikit-image scipy
  cd detectron2
  pip install -e .
  cd projects/DensePose
  pip install -e .

• run in ./detectron2/projects/DensePose:

  CUDA_VISIBLE_DEVICES=0 python apply_net.py get configs/densepose_rcnn_R_50_FPN_s1x.yaml https://dl.fbaipublicfiles.com/densepose/densepose_rcnn_R_50_FPN_s1x/165712039/model_final_162be9.pkl ../../../data/${dir_name}/images --output ../../../data/${dir_name}/images-densepose  -v

  Please check the 'detectron2' repository for details.

• run:

  CUDA_VISIBLE_DEVICES=0 python make_agnostic.py --dir_name ${dir_name} --dil_size ${hair_dilation_size}

Save a text file of test pairs as ./data/${dir_name}/test_pairs.txt

python make_test_pairs.py --dir_path ./data/${dir_name}

• VAE model:
  • download 'realisticVisionV51_v51VAE.ckpt' (3.97G) from https://civitai.com/models/4201?modelVersionId=130072 and save it into ./models
• HairFusion:
  • download and unzip 'hairfusion.zip' (8.4G) and save it into ./logs

bash ./scripts/test.sh

If you find our work useful for your research, please cite us:

@inproceedings{chung2025hairfusion,
  title={What to Preserve and What to Transfer: Faithful, Identity-Preserving Diffusion-based Hairstyle Transfer},
  author={Chung, Chaeyeon and Park, Sunghyun and Kim, Jeongho and Choo, Jaegul},
  booktitle={The Association for the Advancement of Artificial Intelligence},
  year={2025}
}

Licensed under the CC BY-NC-SA 4.0 license https://creativecommons.org/licenses/by-nc-sa/4.0/legalcode.