GPGPU implementation of growing neural gas: application to 3D scene reconstruction

Please use this identifier to cite or link to this item: http://hdl.handle.net/10045/34195
Información del item - Informació de l'item - Item information
Title: GPGPU implementation of growing neural gas: application to 3D scene reconstruction
Authors: Orts-Escolano, Sergio | Garcia-Rodriguez, Jose | Viejo Hernando, Diego | Cazorla, Miguel | Morell, Vicente
Research Group/s: Robótica y Visión Tridimensional (RoViT) | Informática Industrial y Redes de Computadores
Center, Department or Service: Universidad de Alicante. Instituto Universitario de Investigación Informática | Universidad de Alicante. Departamento de Ciencia de la Computación e Inteligencia Artificial | Universidad de Alicante. Departamento de Tecnología Informática y Computación
Keywords: Growing neural gas | Parallel computing | GPU | CUDA | Multicore | 3D reconstruction | Egomotion
Knowledge Area: Ciencia de la Computación e Inteligencia Artificial | Tecnologías de la información y la Computación
Issue Date: Oct-2012
Publisher: Elsevier
Citation: Journal of Parallel and Distributed Computing. 2012, 72(10): 1361-1372. doi:10.1016/j.jpdc.2012.05.008
Abstract: Self-organising neural models have the ability to provide a good representation of the input space. In particular the Growing Neural Gas (GNG) is a suitable model because of its flexibility, rapid adaptation and excellent quality of representation. However, this type of learning is time-consuming, especially for high-dimensional input data. Since real applications often work under time constraints, it is necessary to adapt the learning process in order to complete it in a predefined time. This paper proposes a Graphics Processing Unit (GPU) parallel implementation of the GNG with Compute Unified Device Architecture (CUDA). In contrast to existing algorithms, the proposed GPU implementation allows the acceleration of the learning process keeping a good quality of representation. Comparative experiments using iterative, parallel and hybrid implementations are carried out to demonstrate the effectiveness of CUDA implementation. The results show that GNG learning with the proposed implementation achieves a speed-up of 6× compared with the single-threaded CPU implementation. GPU implementation has also been applied to a real application with time constraints: acceleration of 3D scene reconstruction for egomotion, in order to validate the proposal.
Sponsor: This work has been supported by grant DPI2009-07144 from Ministerio de Ciencia e Innovacion of the Spanish Government, by the University of Alicante projects GRE09-16 and GRE10-35, and Valencian Government project GV/2011/034. Experiments were made possible with a generous donation of hardware from NVDIA.
URI: http://hdl.handle.net/10045/34195
ISSN: 0743-7315 (Print) | 1096-0848 (Online)
DOI: 10.1016/j.jpdc.2012.05.008
Language: eng
Type: info:eu-repo/semantics/article
Peer Review: si
Publisher version: http://dx.doi.org/10.1016/j.jpdc.2012.05.008
Appears in Collections:INV - RoViT - Artículos de Revistas
INV - I2RC - Artículos de Revistas

Files in This Item:
Files in This Item:
File Description SizeFormat 
Thumbnailjdpc.pdfPreprint (acceso abierto)1,7 MBAdobe PDFOpen Preview
Thumbnailjdpc-final.pdfVersión final (acceso restringido)3,11 MBAdobe PDFOpen    Request a copy


Items in RUA are protected by copyright, with all rights reserved, unless otherwise indicated.