3D Point Cloud Denoising via Deep Neural Network based Local Surface Estimation
by Chaojing Duan, Siheng Chen and Jelena Kovacevic
The code is written by Muqiao Yang.
Introduction This repository is for our ICASSP 2019 paper '3D Point Cloud Denoising via Deep Neural Network based Local Surface Estimation'. The code is modified from PointNet and FoldingNet.
Installation This code has been tested with Pytorch 0.4, Python 3.6 and Ubuntu 14+ . We will upload a new version based on Pytorch 1.0 since it is more stable and convenient.
Dataset ShapeNet meshes: https://www.shapenet.org/
ModelNet meshes: http://modelnet.cs.princeton.edu/
You can also train with the dataset in PUNet, which concentrates more on the local geometries: https://github.com/yulequan/PU-Net
Please cite their papers if you use their dataset to train/test.
Experiments
With feature transform
python3.6 utils/train.py --path <dataset path> --filename <file name> --nepoch <num of epochs> --feature_transform
Without feature transform
python3.6 utils/train.py --path <dataset path> --filename <file name> --nepoch <num of epochs>