3D object modeling with graphics hardware acceleration and unsupervised neural networks

Authors:
Felipe Montoya-Franco;Andrés F. Serna-Morales;Flavio Prieto
Affiliations:
Department of Mechanical and Mechatronics Engineering, Universidad Nacional de Colombia, Sede Bogotá, Bogotá, Colombia,;Department of Electrical, Electronic and Computer Engineering, Universidad Nacional de Colombia, Sede Manizales, Manizales, Colombia;Department of Mechanical and Mechatronics Engineering, Universidad Nacional de Colombia, Sede Bogotá, Bogotá, Colombia,
Venue:
ISVC'11 Proceedings of the 7th international conference on Advances in visual computing - Volume Part I
Year:
2011

Citing 6
Cited 0

On the Quantization Error in SOM vs. VQ: A Critical and Systematic Study

WSOM '09 Proceedings of the 7th International Workshop on Advances in Self-Organizing Maps
Parallel ant colony for nonlinear function optimization with graphics hardware acceleration

SMC'09 Proceedings of the 2009 IEEE international conference on Systems, Man and Cybernetics
Research on k-means Clustering Algorithm: An Improved k-means Clustering Algorithm

IITSI '10 Proceedings of the 2010 Third International Symposium on Intelligent Information Technology and Security Informatics
A GPU-based iterated tabu search for solving the quadratic 3-dimensional assignment problem

AICCSA '10 Proceedings of the ACS/IEEE International Conference on Computer Systems and Applications - AICCSA 2010
Spatio-temporal image tracking based on optical flow and clustering: an endoneurosonographic application

MICAI'10 Proceedings of the 9th Mexican international conference on Advances in artificial intelligence: Part I
A GPU Based 3D Object Retrieval Approach Using Spatial Shape Information

ISM '10 Proceedings of the 2010 IEEE International Symposium on Multimedia

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper presents a methodology for reaching higher performances when modeling 3D virtualized reality objects using Self-Organizing Maps (SOM) and Neural Gas Networks (NGN). Our aim is to improve the training speed of unsupervised neural networks when modeling 3D objects using a parallel implementation in a Graphic Process Unit (GPU). Experimental tests were performed over several virtualized reality objects as phantom brain tumors, archaeological items, faces and fruits. In this research, the classic SOM and NGN algorithms were adapted to the data-parallel GPU, and were compared to a similar implementation in an only-CPU platform. We present evidence that rates NGN as a better neural architecture, in quality terms, compared to SOM in the task of 3D object modeling. In order to combine the NGN accuracy with the SOM faster training, we propose and implement a hybrid neural network based on NGN using SOM as seed. Our experimental results show a considerable reduction in the training time without affecting the representation accuracy.