A GPU-Based Framework for Generating Implicit Datasets of Voxelized Polygonal Models for the Training of 3D Convolutional Neural Networks

Carlos J Ogayar-Anguita,Miguel Diaz-Medina,Antonio J Rueda-Ruiz,Angel L Garcia-Fernandez,Rafael J Segura-Sanchez

doi:10.1109/access.2020.2965624

Carlos J Ogayar-Anguita, Miguel Diaz-Medina + Show 3 more

Open Access

PDF Available

https://doi.org/10.1109/access.2020.2965624

Copy DOI

Export

Save

Cite

Journal: IEEE Access	Publication Date: Jan 1, 2020
Citations: 8	License type: CC BY 4.0

Affiliation: University of Jaén

Abstract
Highlights/Summary
Full-Text PDF
Similar Papers

Abstract

Listen

In this paper we present an efficient GPU-based framework to dynamically perform the voxelization of polygonal models for training 3D convolutional neural networks. It is designed to manage the dataset augmentation by using efficient geometric transformations and random vertex displacements in GPU. With the proposed system, every voxelization is carried out on-the-fly for directly feeding the network. The computing performance with this approach is much better than with the standard method, that carries out every voxelization of each model separately and has much higher data processing overhead. The core voxelization algorithm works by using the standard rendering pipeline of the GPU. It generates binary voxels for both the interior and the surface of the models. Moreover, it is simple, concise and very compatible with commodity hardware, as it neither uses complex data structures nor needs vendor-specific hardware or additional dependencies. This framework dramatically reduces the input/output operations, and completely eliminates the storage footprint of the voxelization dataset, managing it as an implicit dataset.

Highlights

In this paper we present an efficient framework for dynamically performing the voxelization of B-rep models based on polygonal meshes
The GPU-based algorithm used for voxelizing 3D models is concise, very compatible with commodity hardware, and it neither uses complex data structures nor has additional dependencies such as CUDA
We have implemented a dummy 3D Convolutional Neural Network (3D-CNN) used through its C++ interface for testing the performance of the framework

Summary

Introduction

One of the most promising methods is the 3D Convolutional Neural Network (3D-CNN), which relies on a volumetric representation of the data. Several methods based on 3D-CNNs have been recently proposed for the purposes of object classification [1]–[9]. Machine learning methods heavily rely on the availability of a large amount of data, so a scarce dataset consisting of real-world objects Voxel representations of 3D models have many applications in solid modeling, volume graphics and physical simulation. They have been extensively used for rendering objects which are difficult to represent with traditional surface representations, like clouds, fire, smoke or terrain models [13].

Methods

Results

Conclusion