PolyGNN

---

PolyGNN is an implementation of the paper PolyGNN: Polyhedron-based Graph Neural Network for 3D Building Reconstruction from Point Clouds.

Note

This repository is undergoing revisions and may differ from the state of the arXiv manuscript.

Dependencies

All-in-one installation

Create a conda environment with all dependencies installed:

git clone https://github.com/chenzhaiyu/polygnn && cd polygnn
conda env create -f environment.yml && conda activate polygnn

Manual installation

Create a conda environment and enter it:

conda create --name polygnn python=3.10 && conda activate polygnn

Install mamba for faster package parsing and installation:

conda install mamba -c conda-forge

Install the main dependencies:

mamba install pytorch torchvision sage=10.0 pytorch-cuda=11.7 pyg=2.3 pytorch-scatter pytorch-sparse pytorch-cluster torchmetrics rtree -c pyg -c pytorch -c nvidia -c conda-forge
pip install abspy hydra-core hydra-colorlog omegaconf trimesh tqdm wandb plyfile

Usage

Download data and weights:

python download.py dataset=munich

Train PolyGNN:

python train.py dataset=munich

Evaluate PolyGNN:

python test.py dataset=munich evaluate.save=true

Reconstruct meshes from predictions:

python reconstruct.py dataset=munich reconstruct.type=mesh

Remap meshes to original CRS:

python remap.py dataset=munich

Generate statistics:

python stats.py dataset=munich

TODOs

• Host data and weights
• Short tutorial on getting started
• Scripts for data generation and manipulation

License

MIT

Citation

If you use PolyGNN in a scientific work, please consider citing the paper:

@article{chen2023polygnn,
  title={PolyGNN: polyhedron-based graph neural network for 3D building reconstruction from point clouds},
  author={Chen, Zhaiyu and Shi, Yilei and Nan, Liangliang and Xiong, Zhitong and Zhu, Xiao Xiang},
  journal={arXiv preprint arXiv:2307.08636},
  year={2023}
}