Colin97 / Msn Point Cloud Completion

MSN: Morphing and Sampling Network for Dense Point Cloud Completion

[paper] [data]

MSN is a learning-based shape completion method which can preserve the known structures and generate dense and evenly distributed point clouds. See our AAAI 2020 paper for more details.

In this project, we also provide an implementation for the Earth Mover's Distance (EMD) of point clouds, which is based on the auction algorithm and only needs $O(n)$ memory.

with 32,768 points after completion

Usage

1) Envrionment & prerequisites

• Pytorch 1.2.0
• CUDA 10.0
• Python 3.7
• Visdom
• Open3D

2) Compile

Compile our extension modules:

cd emd
python3 setup.py install
cd expansion_penalty
python3 setup.py install
cd MDS
python3 setup.py install

3) Download data and trained models

Download the data and trained models from here. We don't provide the partial point clouds of the training set due to the large size. If you want to train the model, you can generate them with the code and ShapeNetCore.v1. We generate 50 partial point clouds for each CAD model.

4) Train or validate

Run python3 val.py to validate the model or python3 train.py to train the model from scratch.

EMD

We provide an EMD implementation for point cloud comparison, which only needs $O(n)$ memory and thus enables dense point clouds (with 10,000 points or over) and large batch size. It is based on an approximated algorithm (auction algorithm) and cannot guarantee a (but near) bijection assignment. It employs a parameter $\epsilon$ to balance the error rate and the speed of convergence. Smaller $\epsilon$ achieves more accurate results, but needs a longer time for convergence. The time complexity is $O(n^2k)$, where $k$ is the number of iterations. We set a $\epsilon = 0.005, k = 50$ during training and a $\epsilon = 0.002, k = 10000$ during testing. Please refer toemd/README.md for more details.

Citation

If you find our work useful for your research, please cite:

@article{liu2019morphing,
  title={Morphing and Sampling Network for Dense Point Cloud Completion},
  author={Liu, Minghua and Sheng, Lu and Yang, Sheng and Shao, Jing and Hu, Shi-Min},
  journal={arXiv preprint arXiv:1912.00280},
  year={2019}
}

License

This project Code is released under the Apache License 2.0 (refer to the LICENSE file for details).