StyleHairEdit: Advancing Hair Transfer and Editing through Refined Mask Completion and Vibrant Appearance Preservation

Abstract StyleHairEdit advances the field of realistic hairstyle manipulation through an innovative GAN-inversion approach. While recent deep generative models have made significant strides in hair editing, challenges persist in accurately reflecting user preferences and reconstructing complex hair features. Our system addresses these issues by introducing two key improvements: refined mask completion for enhanced structure blending, and a well-designed loss function for vibrant appearance reconstuction. The refined mask completion technique significantly improves the semantic alignment between source and reference images, enabling more natural hair transfer. Our proposed loss function, incorporating masked perceptual and style components, effectively captures and reconstructs detailed hair features, including vivid colors and intricate strain textures. Additionally, we introduce user-controllable components that seamlessly integrate with our enhanced hair transfer process, allowing for intuitive and precise editing. Our system not only excels in hair transfer but also provides a user-friendly sketch interface for further refinement, leveraging the improved transfer capabilities. Quantitative and qualitative evaluations, including user preference studies, demonstrate that StyleHairEdit outperforms current state-of-the-art techniques in both hairstyle generation quality and usability, marking a significant advancement in GAN-based hair manipulation.

Description

Official Implementation of "StyleHairEdit" (under review). KEEP UPDATING! Please Git Pull the latest version.

Updates

2024/06/07 All source codes have been uploaded

Installation

• System requirement: Ubuntu22.04, Windows 11, Cuda 12.1

• Tested GPUs: RTX4090

• Dependencies:
  We recommend running this repository using Anaconda. All dependencies for defining the environment are provided in environment.yaml.

• Create conda environment:

conda create -n HairTrans python=3.9
conda activate HairTrans

• Clone the repository:

git clone https://github.com/korfriend/VividHairStyler.git
cd VividHairStyler

• Install packages with pip:

pip install -r requirements.txt

Download sample images

Please download the FFHQ and put them in the /${PROJECT_ROOT}/database/ffhq directory.

Getting Started

Prerequisites

$ pip install pytorch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 pytorch-cuda=12.1 -c pytorch -c nvidia
$ pip install --upgrade diffusers[torch]

Download pretrained models

Clone the the pretrained models into /${PROJECT_ROOT}/pretrained_models directory :

Model	Description
FFHQ StyleGAN	StyleGAN model pretrained on FFHQ with 1024x1024 output resolution. This includes ffhq_PCA.npz and ffhq.pt, which will be automatically downloaded.
Face Parser Model (BiSeNet)	Pretrained face parse model taken from Barbershop. This model file is seg.pth, which will be automatically downloaded.
Face Landmark Model	Used to align unprocessed images.
FFHQ Inversion Model	Pretrained image embedding model taken from encoder4editing. This model file is e4e_ffhq_encode.pt, which will be automatically downloaded.
Sketch2Image Model	Pretrained sketch hair model taken from SketchHairSalon. This includes 400_net_D.pth, 400_net_G.pth for S2I_braid, 200_net_D.pth, 200_net_G.pth for S2I_unbraid, and 200_net_D.pth, 200_net_G.pth for S2M. These pretrained files need to be manually downloaded and placed in /${PROJECT_ROOT}/pretrained_models.
HairMapper	Pretrained removing hair model taken from HairMapper (You can get it by filling out their Google form for pre-trained models access). This model file is best_model.pt located in the final folder. This pretrained file needs to be manually downloaded and placed in /${PROJECT_ROOT}/pretrained_models.

Model Organization

The pretrained models should be organized as follows:

./pretrained_models/
├── e4e_ffhq_encode.pt (will be downloaded when running)
├── ffhq_PCA.npz (will be downloaded when running)
├── ffhq.pt (will be downloaded when running)
├── final
│   └── best_model.pt 
├── S2I_braid
│   ├── 400_net_D.pth 
│   └── 400_net_G.pth 
├── S2I_unbraid
│   ├── 200_net_D.pth 
│   └── 200_net_G.pth 
├── S2M
│   ├── 200_net_D.pth 
│   └── 200_net_G.pth 
└── seg.pth (will be downloaded when running)

Web UI

You can use the web UI by running the following command in the /VividHairStyler directory:

streamlit run VividHairStyler.py

Acknowledgments

This code borrows heavily from BARBERSHOP.