CUDA + OpenCV

[Daniel Cabrini Hauagge]

Hello everybody,

Some time ago there was a discussion about implementing vision
algorithms on the GPU using CUDA. I just finished implementing the
algorithm described by

Bayesian Multi-scale Differential Optical Flow
E P Simoncelli, Handbook of Computer Vision and Applications, pages
397--422. Academic Press, Apr 1999.

using CUDA and OpenCV.

For pyramids of height 4 and 640x480 images the implementation runs at
90FPS, although I don't have CPU implementation right here with me I
did try one some
years ago on a Pentium 4@3GHz and got 1FPS with 380x240 images. I
didn't try to optimize the code yet.

If anybody is interested in having a look at it here's the link
http://www.liv.ic.unicamp.br/~hauagge/files/CUDABOF.zip

--
Daniel Cabrini Hauagge
cel: [phone]

-------------------------------------------------------------------------
Check out the new SourceForge.net Marketplace.
It's the best place to buy or sell services for
just about anything Open Source.
http://sourceforge.net/services/buy/index.php
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

[Stefan Hafeneger]

Hi Daniel & all,

I just arrived back home from my trip to this years WWDC, Apple's  
Developer Conference. Concerning CUDA (I made the suggestion thinking  
of using CUDA in OpenCV), Apple just announced OpenCL (Open Computing  
Library) that will be available in the next release of OS X Snow  
Leopard and is a potential successor of CUDA and all other hardware  
specific parallel programming languages. It also uses a superset of C  
but uses both GPUs and CPUs and is platform independent. So if you are  
running a 8-core PowerMac with a few strong graphic cards inside you  
get a really nice performance. I'm not sure if this will be really  
open source, but Apple currently tries to establish it as a standard  
in the industry. So perhaps it's better to concentrate on OpenCL  
instead of CUDA and the other platform dependent solutions for the  
future versions of OpenCV.

With best wishes, Stefan


Am 16.06.2008 um 01:35 schrieb Daniel Cabrini Hauagge:

> Hello everybody,
>
> Some time ago there was a discussion about implementing vision
> algorithms on the GPU using CUDA. I just finished implementing the
> algorithm described by
>
> Bayesian Multi-scale Differential Optical Flow
> E P Simoncelli, Handbook of Computer Vision and Applications, pages
> 397--422. Academic Press, Apr 1999.
>
> using CUDA and OpenCV.
>
> For pyramids of height 4 and 640x480 images the implementation runs at
> 90FPS, although I don't have CPU implementation right here with me I
> did try one some
> years ago on a Pentium 4@3GHz and got 1FPS with 380x240 images. I
> didn't try to optimize the code yet.
>
> If anybody is interested in having a look at it here's the link
> http://www.liv.ic.unicamp.br/~hauagge/files/CUDABOF.zip
>
> --
> Daniel Cabrini Hauagge
> cel: [phone]
>
> -------------------------------------------------------------------------
> Check out the new SourceForge.net Marketplace.
> It's the best place to buy or sell services for
> just about anything Open Source.
> http://sourceforge.net/services/buy/index.php
> _______________________________________________
> Opencvlibrary-devel mailing list
> Opencvlibrary-devel@...
> https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

-------------------------------------------------------------------------
Check out the new SourceForge.net Marketplace.
It's the best place to buy or sell services for
just about anything Open Source.
http://sourceforge.net/services/buy/index.php
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel