Self-Sampling - for neural point cloud consolidation

We introduce a novel technique for neural point cloud consolidation which learns from only the input point cloud.

TOG 2021 [Paper] [Project Page]

by Gal Metzer, Rana Hanocka, Raja Giryes, and Daniel Cohen-Or

Getting Started

Installation

Clone this repo:

Setup Conda Environment

Relies on PyTorch version 1.7.1
Pytorch Geometric
Everything can be installed via conda environment conda env create -f env.yml (creates an environment called self-sample)

Running Examples

The demos folder contains examples from the paper.
For each shape the demo runs the optimization and inference parts.
For instance, to run the lamp demo simply execute from the root project folder:

demos/lamp.sh

The results would be found at demos-results/lamp/lamp_result.xyz,
and respectively for the other shapes as well.

Example shapes

alien, anchor, lamp - sharp point consolidation
candle, scanned Leg, tiki - sparse point consolidation
camera_noised - denoising

Citation

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

@article{metzer2020self,
author = {Metzer, Gal and Hanocka, Rana and Giryes, Raja and Cohen-Or, Daniel},
title = {Self-Sampling for Neural Point Cloud Consolidation},
year = {2021},
issue_date = {October 2021},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
volume = {40},
number = {5},
issn = {0730-0301},
url = {https://doi.org/10.1145/3470645},
doi = {10.1145/3470645},
}

Questions / Issues

If you have questions or issues running this code, please open an issue.

Note: the original implementation used this implementation of PointNet++, which is not guaranteed to supported newer versions of pytorch.
This implementation uses Pytorch Geometric instead, which can not hold large subsets at train time.

Therefore, demos are designed for subset sizes lower than used in the paper. Increasing the subset size to 12K-14K on an appropriate GPU, improves the accuracy of the results.

papiamahato/self-sample