Computer Graphics Laboratory ETH Zurich

ETH

Geometry Representations with Unsupervised Feature Learning

Y. J. Yoon, A. Lelidis, A. C. Oztireli, J. M. Hwang, M. Gross, S. M. Choi

Proceedings of 2016 International Conference on Big Data and Smart Computing (BigComp) (Hong Kong, China, January 8-10, 2016), pp. 18-20
[Abstract] [BibTeX] [PDF]

Abstract

Geometry data in massive amounts can be generated thanks to the modern capture devices and mature geometry modeling tools. It is essential to develop the tools to analyze and utilize this big data. In this paper, we present an exploration of analyzing geometries via learning local geometry features. After extracting local geometry patches, we parameterize each patch geometry by a radial basis function based interpolation. We use the resulting coefficients as discrete representations of the patches. These are then fed into feature learning algorithms to extract the dominant components explaining the overall patch database. This simple approach allows us to handle general representations such as point clouds or meshes with noise, outliers, and missing data. We present features learned on several patch databases to illustrate the utility of such an analysis for geometry processing applications.

@INPROCEEDINGS{7425812,
author={Yeo-Jin Yoon and A. Lelidis and A. C. Öztireli and Jung-Min Hwang and M. Gross and Soo-Mi Choi},
booktitle={2016 International Conference on Big Data and Smart Computing (BigComp)},
title={Geometry representations with unsupervised feature learning},
year={2016},
pages={137-142},
keywords={Big Data;computer graphics;geometry;interpolation;radial basis function networks;unsupervised learning;3D geometry data;big geometry data;geometry representations;interpolation;local geometry patches extraction;patch geometry;radial basis function;unsupervised feature learning;Atomic measurements;Dictionaries;Feature extraction;Geometry;Image reconstruction;Shape;Three-dimensional displays;Geometry representations;big geometry data;dictionary learning},
doi={10.1109/BIGCOMP.2016.7425812},
month={Jan},}
[Download BibTeX]

Overview

Geometry data in massive amounts can be generated thanks to the modern capture devices and mature geometry modeling tools. It is essential to develop the tools to analyze and utilize this big data. In this paper, we present an exploration of analyzing geometries via learning local geometry features. After extracting local geometry patches, we parameterize each patch geometry by a radial basis function based interpolation. We use the resulting coefficients as discrete representations of the patches. These are then fed into feature learning algorithms to extract the dominant components explaining the overall patch database. This simple approach allows us to handle general representations such as point clouds or meshes with noise, outliers, and missing data. We present features learned on several patch databases to illustrate the utility of such an analysis for geometry processing applications.

Downloads

Download Paper
[PDF]