12th International Workshop on Image Analysis for Multimedia Interactive Services (WIAMIS 2011), Delft, The Netherlands, April 2011. Kompatsiaris, 'GPU Acceleration for Support Vector Machines', Proc. First release of the GPU-Accelerated LIBSVM in Windows 32/64 bit.Ī first document describing some of the work related to the GPU-Accelerated LIBSVM is the following please cite it if you find this implementation useful in your work:Ī.Video memory allocation is checked, producing an error when memory limits are exceeded.Using CUBLAS_V2 which is compatible with the CUDA SDK v4.0 and up.Windows 32-bit version, Windows 64-bit version. Windows 32-bit version, Windows 64-bit version, source code (VS2008 project). This version is based on LIBSVM version 3.0 & CUDA SDK 3.0.Įxecutables and code can be downloaded here: Source code: Windows Visual Studio 2010 project, Linux makefile. The latest version is based on LIBSVM version 3.17 & CUDA SDK 5.5.Įxecutables: Windows 32-bit, Windows 64-bit, Linux 64-bit. Please take into consideration input data size limitations that can occur from the memory capacity of the graphics card that is used.This gain is increasing dramatically with the size of the dataset. GPU-accelerated LIBSVM gives a performance gain depending on the size of input data set.Parameter optimization using the easy.py script provided by LIBSVM.20 different feature models with a variable number of input training vectors ranging from 36 up to 3772.Training vectors with a dimension of 6000.TRECVID 2007 Dataset for the detection of high level features in video shots.To showcase the performance gain using the GPU-accelerated LIBSVM we present an example run. Watch a short video on the capabilities of the GPU-accelerated LIBSVM package here The modifications were done in the kernel computation, that is now performed using the GPU. The functionality and interface of LIBSVM remains the same. GPU-accelerated LIBSVM is a modification of the original LIBSVM that exploits the CUDA framework to significantly reduce processing time while producing identical results. LIBSVM Accelerated with GPU using the CUDA Framework

Libsvm download