README.md

SSP: Semi-signed prioritized neural fitting for surface reconstruction from unoriented point clouds (WACV2024)

Page, Paper, Poster

Runsong Zhu¹*, Di Kang², Ka-Hei Hui¹, Yue Qian², Shi Qiu¹, Zhen Dong³, Linchao Bao², Pheng-Ann Heng¹, Chi-Wing Fu¹.

(*Work partially done during an internship at Tencent AI Lab)

¹The Chinese University of Hong Kong + ²Tencent AI Lab + ³Wuhan University.

Set up

pip install -r code/reconstruction/requirements.txt

How to use the code

Data praparation (calculating the outside regions).

cd code/space_carving
python generate_abc_adapative_add_outside_depth.py
python generate_pcp_adaptive_add_outside_depth.py
python generate_SRB_adaptive_add_outside_depth.py

We also provide the processed data (Here). Note that, we use the AdaFit to calculate the unoriented normals to boost the performance on mentioned datasets in our paper. If you want to test your data, you could generate the normals using AdaFit or other methods (e.g., PCA).

Start fitting for given point clouds input

cd code/reconstruction
python SSP.py --dataset thingi  --shape 120477 --nepoch 10000 --outdir thingi10K_exps # example

Acknowledgements

Some code snippets are borrowed from IGR and SAP codebases. We thank the authors for releasing their code.

Citation

If you find our work useful in your research, please cite our paper.

@inproceedings{zhu2024ssp,
  title={SSP: Semi-Signed Prioritized Neural Fitting for Surface Reconstruction From Unoriented Point Clouds},
  author={Zhu, Runsong and Kang, Di and Hui, Ka-Hei and Qian, Yue and Qiu, Shi and Dong, Zhen and Bao, Linchao and Heng, Pheng-Ann and Fu, Chi-Wing},
  booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
  pages={3769--3778},
  year={2024}
}