Learning Signed Distance Functions from Noisy 3D Point Clouds via Noise to Noise Mapping (ICML 2023 Oral)

Project Page | Personal Web Page | Paper

This repository contains the code to reproduce the results from the paper. Learning Signed Distance Functions from Noisy 3D Point Clouds via Noise to Noise Mapping

You can find detailed usage instructions for preprocessing data and training your own models below.

Installation

First you have to make sure that you have all dependencies in place. The simplest way to do so, is to use anaconda.

You can create an anaconda environment called tf using

conda env create -f tf.yaml
conda activate tf

Data Processing

First downloading the dataset you would like to use (e.g. ShapeNet). Then you can generate the noisy point cloud data for training as:

python gen_noise.py /PATH/TO/YOUR/DATA/dataname.ply(obj,off,etc.) /PATH/TO/SAVEDATA/savename.ply

You can simply modify the code for processing your own data.

Training

You can now train the network by runing

python noise2noise.py --dataname savename --data_dir /PATH/TO/SAVEDATA --CUDA 0 --out_dir /PATH/TO/OUTPUT --train --save_idx -1

Related work

Pytorch 
https://github.com/mabaorui/NeuralPull-Pytorch
https://github.com/junshengzhou/CAP-UDF
Tensorflow
https://github.com/mabaorui/NeuralPull
https://github.com/mabaorui/OnSurfacePrior
https://github.com/mabaorui/PredictableContextPrior

Citation

If you find our code or paper useful, please consider citing

@inproceedings{BaoruiNoise2NoiseMapping,
    title = {Learning Signed Distance Functions from Noisy 3D Point Clouds via Noise to Noise Mapping},
    author = {Baorui Ma and Yu-Shen Liu and Zhizhong Han},
    booktitle = {International Conference on Machine Learning (ICML)},
    year = {2023}
}

Surface Reconstruction Demo