Transforming Unstructured Hair Strands into Procedural Hair Grooms

This repository contains the implementation for the SIGGRAPH 2025 paper Transforming Unstructured Hair Strands into Procedural Hair Grooms.

This code has been tested on Ubuntu 22.04 with an RTX 4090 GPU. It requires an NVIDIA GPU with CUDA to run.

Install dependencies for nvdiffrast:

./setup_dependencies.sh

Using conda,

conda env create --file environment.yml
conda activate inverse_hair_grooming

Place examples in the data directory.

Run an example

python train-3d.py --dataset [example]

This will save to output/[example]/[experiment number] results including:

• Reference and final visualizations reference.png and final.png
• Checkpoints stagen_*.pt, final.pt and visualizations stagen_*.png
• Loss and parameter plots training.png

Note: When running this project in a docker container on x86_64, you might encounter a nvdiffrast compilation error because the file crt/host_config.h cannot be found. This file and other missing dependencies are located in $CONDA_ROOT/targets/x86_64-linux/include instead of $CONDA_ROOT/include. We have found that copying or symlinking the missing files to $CONDA_ROOT/include fixes the issue.

1. Run Gaussian Haircut (GH): https://github.com/eth-ait/GaussianHaircut
2. Run the script to create the example:

   python scripts/create_example.py --type gh --in_dir [GH output]/[example name] --example [example]

3. Edit the example scene-3d.py file in data/[example]_gh.

1. Get MonoHair (MH) data: https://github.com/KeyuWu-CS/MonoHair
2. Run the script to create the example:

   python scripts/create_example.py --type mh --in_dir [MH output]/[example name] --example [example]

3. Edit the example scene-3d.py file in data/[example]_mh

Render results using Blender:

python scripts/render.py --blender_path [blender path] --dataset [example] --exp_num [experiment number] --view [view number]

This will save rendered images to output/[example]/[experiment number]/images.

Run synthetic-grooms-coily first, then see example in scripts/editing.py.

python train-3d.py --dataset synthetic-grooms-coily
python scripts/editing.py

This will save rendered images to output/synthetic-grooms-coily/[last experiment number]/images.

This project is under the "CC BY-NC 4.0" license, which terms can be found in the LICENSE file.

@inproceedings{Chang2025IPHG,
  author = {Chang, Wesley and Russell, Andrew L. and Grabli, Stephane and Chiang, Matt Jen-Yuan and Hery, Christophe and Roble, Doug and Ramamoorthi, Ravi and Li, Tzu-Mao and Maury, Olivier},
  title = {Transforming Unstructured Hair Strands into Procedural Hair Grooms},
  year = {2025},
  issue_date = {August 2025},
  publisher = {Association for Computing Machinery},
  address = {New York, NY, USA},
  volume = {44},
  number = {4},
  url = {https://doi.org/10.1145/3731168},
  doi = {10.1145/3731168},
  journal = {ACM Trans. Graph.},
  month = aug,
  numpages = {20}
}