Yuming Zhao · Qijian Zhang · Junhui Hou · Jiazhi Xia · Wenping Wang · Ying He

This repository contains the official implementation for the paper FlexPara: Flexible Neural Surface Parameterization.

We have previously conducted a series of works on regular 3D geometry representations. Please refer to the following:

• FlattenAnything for global free-boundary surface parameterization.
• RegGeoNet for large-scale 3D point clouds.
• Flattening-Net for feed-forward point cloud structurization.
• SPCV for dynamic 3D point cloud sequences.

conda create --name FlexPara python=3.9
conda activate FlexPara
pip install torch==1.10.1+cu111 torchvision==0.11.2+cu111 torchaudio==0.10.1 -f https://download.pytorch.org/whl/cu111/torch_stable.html
pip install -r requirements.txt

cd cdbs/CD
python setup.py install

cd ..
cd EMD
python setup.py install
cp build/lib.linux-x86_64-cpython-39/emd_cuda.cpython-39-x86_64-linux-gnu.so .

The project is a v0.1 version for fast review now, and we will release the v1.0 version later, including data pre-processing, full evaluation and so on.

Training:

mkdir expt
cd scripts
# Global Parameterization
python train.py 1 ../data/bunny.obj ../expt 1600 10000
# MulitChart Parameterization
python train.py 8 ../data/bunny.obj ../expt 1600 10000

Testing:

mkdir expt
cd scripts
# Global Parameterization
python test.py ../data/bunny.obj flexpara_global.pth ../expt
# MulitChart Parameterization
python test.py ../data/bunny.obj flexpara_multi_8.pth ../expt

• data pre-processing
• environment configuration
• train code
• test code (simple version)
• test code (full version)

If you find our work useful in your research, please consider citing:

@article{zhao2025flexpara,
  title={FlexPara: Flexible Neural Surface Parameterization},
  author={Zhao, Yuming and Zhang, Qijian and Hou, Junhui and Xia, Jiazhi and Wang, Wenping and He, Ying},
  journal={arXiv preprint arXiv:2504.19210},
  year={2025}
}