Vision library using CUDA?

Accelerated Computing CUDA CUDA Programming and Performance
1

Dear all,

Is there anybody who knows about any library that runs on Linux and utilises GPUs via CUDA framework please?

I had a look and found OpenVIDIA,GPUCV and GPU4VISION. OpenVIDIA does not reply to any messages neither on their forum nor via email and I cannot get their code working.
I am not sure if the GPUCV project is still active or not, have not received any replies to my inquiries and can’t compile the code either. GPU4VISION looks promising but it seems that the only library that is available on Linux is for optic flow calculations only.

Please correct me if I am wrong. I hope I am wrong, since GPUs deserve great and well supported libraries for vision processing :D

Kind Regards,
Martin Peniak

2

Did you look into OpenCV, They work on some GPU support now.

3

There are many open-source libraries out there that are based around Cuda and OpenCV. For example:

CUDA SURF: Max-Planck-Institut für Informatik: Publications
Affine-Invariant KLT: IMU-aided Affine-photometric KLT Feature Tracker in CUDA Framework
CudaCuts: http://cvit.iiit.ac.in/resources/cudacuts/cudaCuts.zip

Like philippev said, OpenCV is also adding cuda support directly. Currently they have stereo disparity and surf I think.

4

Thanks guys for your replies. I have in fact integrated OpenCV into my application. A good integration of OpenGL, OpenCV, CUDA and Qt has increased the performance of image processing for our humanoid robot iCub.
This is implemented in Aquila, a software toolkit designed primarily for developmental robotics community: http://www.martinpeniak.com/index.php?option=com_content&view=article&id=170&Itemid=128

Best Regards,
Martin Peniak

5

The MTRNN looks cool, very interesting neural network. 1000x - that’s a nice speedup for cuda External Image