STAR-Edge (CVPR‘25)

In this work, we presented STAR-Edge, a novel point cloud edge extraction approach specifically designed for thin-walled structures. The core of our method is the introduction of a new structure-aware neighborhood representation—the local spherical curve. This representation effectively emphasizes co-planar points while minimizing interference from unrelated surfaces, enabling robust and accurate edge detection in complex thin-walled geometries.

Preparation

Installation Prerequisites

• Python 3.8
• CUDA 11.8
• PyTorch 2.1.2
• tqdm, natsort

pip install tqdm
pip install natsort

• fftw

wget http://www.fftw.org/fftw-3.3.9.tar.gz
tar -xvf fftw-3.3.9.tar.gz
cd fftw-3.3.9
mkdir build
cd build
cmake ..
make install

You may need to add the library path

export LD_LIBRARY_PATH=/usr/local/lib:$LD_LIBRARY_PATH

• LocalSH

cd LocalSH
mkdir build
cd build
cmake ..
make install

Demo

python eval_all.py

Data Preparation

• The data is expected to be in the following file structure:

  |-- dataset/
      |-- thin_walled/
          |-- gt
          |-- pc
  |-- dataset_pre/
      |-- dataset_pre_thin/
          |-- cls_gt
          |-- desc
          |-- pc_normal
  
  

Usage

Train

python train_thin.py

Test

python eval_thin.py

Visualization Results

Visual comparison on the thin-walled structure dataset.

Visual comparison on 3D shapes from the ABC dataset.

Visualization on real thin-walled structure data.

Citation

@article{li2025star,
  title={STAR-Edge: Structure-aware Local Spherical Curve Representation for Thin-walled Edge Extraction from Unstructured Point Clouds},
  author={Li, Zikuan and Chen, Honghua and Wang, Yuecheng and Wu, Sibo and Wei, Mingqiang and Wang, Jun},
  journal={arXiv preprint arXiv:2503.00801},
  year={2025}
}

Acknowledgements

Our code is inspired by fftw.