GMM based Fisher vector calculation on GPGPU
Data Mining and Web search Research Group, Informatics Laboratory, Computer and Automation Research Institute of the Hungarian Academy of Sciences
Computer and Automation Research Institute of the Hungarian Academy of Sciences, 2011
@article{petras2011gmm,
title={GMM based Fisher vector calculation on GPGPU},
author={Petr{‘a}s, E.B.B.D.I. and Bencz{‘u}r, A.A.},
year={2011}
}
We describe an accurate yet very fast implementation of a visual word generation method by using general purpose graphical processors (GPUs). Visual words have recently proved to be a key tool in image classification. Best performing Pascal VOC and ImageCLEF systems use Gaussian mixtures or k-means clustering to define visual words based on the content-based features of points of interest. In many cases, Gaussian Mixture Modeling (GMM) is more accurate but computationally expensive compared to other methods, sometimes taking days to compute for the standard research tasks. We reach a 14-times speedup over a well-tuned sequential GMM implementation. We measure the accuracy of our methods over the Pascal VOC 2007 and give results comparable to the best teams with reduced computational time. Since most image processing components already have GPU implementations, we believe our results make large scale image classification with Fisher vectors scalable with the help of graphics processors.
December 5, 2011 by hgpu
