Awesome

FPNN: Field Probing Neural Networks for 3D Data

Created by <a href="http://web.stanford.edu/~yangyan/" target="_blank">Yangyan Li</a>, <a href="http://www.pirk.info/" target="_blank">Soeren Pirk</a>, <a href="http://ai.stanford.edu/~haosu/" target="_blank">Hao Su</a>, <a href="http://web.stanford.edu/~rqi/" target="_blank">Charles Ruizhongtai Qi</a>, and <a href="http://geometry.stanford.edu/member/guibas/" target="_blank">Leonidas J. Guibas</a> from Stanford University.

Introduction

We propose a light-weight way for learning features from 3D data. See more details from our <a href="http://arxiv.org/abs/1605.06240" target="_blank">research paper on arXiv</a> (was accepted to NIPS 2016).

Usage

Check <a href="https://github.com/yangyanli/FPNN/tree/master/training_settings" target="_blank">training settings</a> for example usage of the field probing layers, as well as logs generated during our training.

From FPNN to PointCNN

If you are interested in FPNN, we highly recommend you take a look at PointCNN, which outperforms FPNN in terms of ModelNet40 classification, together with other advantages.