MODNet: Multi-offset Point Cloud Denoising Network Customized for Multi-scale Patches(PG2022)

This is our implementation of MODNet, a network that employs multi-scale geometric perception information to guide the network to utilize multi-scale information.

Environment

• Python 3.6
• PyTorch 1.5.0
• CUDA and CuDNN (CUDA 10.1 & CuDNN 7.5)
• TensorboardX (2.0) if logging training info.

Datasets

You can download the datasets used in this work from the following train, test, and copy the decompressed Train and Test folders to ./Dataset.

Setup

Install required python packages:

pip install numpy
pip install scipy
pip install plyfile
pip install scikit-learn
pip install tensorboardX (only for training stage)
pip install torch==1.5.0+cu101 torchvision==0.6.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html

Clone this repository:

git clone https://https://github.com/hay-001/MODNet.git
cd MODNet

Compile Chamfer Opertors (only for evaluation)

cd ./MODNet/Customer_Module/chamfer_distance
python setup.py install

Train

Use the script train.py to train a model in the our dataset (the re-trained model will be saved at ./Summary/Train):

cd MODNet
python train.py

Test with Pre-trained Model

You can download the Pre-trained Model from the following Pre-trained model, and you should copy the Pre-trained model model_full_ae.pth to./Summary/pre_train_model. Use the script test.py to test our test dataset (the results will be saved at ./Summary/Test):

cd MODNet
python test.py

Test with Re-trained Model

Before running the test.py, you should copy the re-trained model model_full_ae_x.pth to ./Summary/pre_train_model and rename it to model_full_ae.pth. Use the script test.py to test our test dataset (the results will be saved at ./Summary/Test):

cd MODNet
python test.py

Evaluation（CD、MSE）

Before running the eval.py, the results should be saved at ./Summary/Test:

cd MODNet
python eval.py

Evaluation（P2M）

Before running the eval_p2m.py, the results should be saved at ./Summary/Test. You may need to recreate a envs(see：https://github.com/luost26/score-denoise) to run this script eval_p2m.py.

cd MODNet
python eval_p2m.py

Citation

If you use our code, please consider citing our paper:

@article {10.1111:cgf.14661,
journal = {Computer Graphics Forum},
title = {{MODNet: Multi-offset Point Cloud Denoising Network Customized for Multi-scale Patches}},
author = {Huang, Anyi and Xie, Qian and Wang, Zhoutao and Lu, Dening and Wei, Mingqiang and Wang, Jun},
year = {2022},
publisher = {The Eurographics Association and John Wiley & Sons Ltd.},
ISSN = {1467-8659},
DOI = {10.1111/cgf.14661}
}

Acknowledgements

This code largely benefits from following repositories:

• PointNet
• PCPNet
• Pointfilter
• score-denoise