Imaging-Processing library with device functions

Hi,

I am searching for a cuda library that will provide device functions for Image analysis/processing. I have found many libraries written in cuda, but they don’t provide device functions, only C functions that make the global calls. I need the interface to provide device functions because I want to do all the processing in the GPU, saving the transfer times between host and device. Does anybody know one?

Thanks,
Cristobal

global calls will leave the data on the device if you are passing device pointers as arguments.

global calls will leave the data on the device if you are passing device pointers as arguments.

I understand. Thanks a lot.

I understand. Thanks a lot.

I understand. Thanks a lot.

What algorithms/functions are you particularly interested in? You can try out CUVI Lib (CUDA Vision and Imaging Library). It does not expose device functions or kernels but the wrappers and the only additional thing that wrappers do is to select optimized execution configuration (block size, grids etc) and launch the kernel (or a series of kernels). Memory transfers are also handled by these wrapper functions exposed through the CUVI API (see cuvi.h header file).

If you can specify your requirement then maybe we’ll be able to help.

What algorithms/functions are you particularly interested in? You can try out CUVI Lib (CUDA Vision and Imaging Library). It does not expose device functions or kernels but the wrappers and the only additional thing that wrappers do is to select optimized execution configuration (block size, grids etc) and launch the kernel (or a series of kernels). Memory transfers are also handled by these wrapper functions exposed through the CUVI API (see cuvi.h header file).

If you can specify your requirement then maybe we’ll be able to help.

What algorithms/functions are you particularly interested in? You can try out CUVI Lib (CUDA Vision and Imaging Library). It does not expose device functions or kernels but the wrappers and the only additional thing that wrappers do is to select optimized execution configuration (block size, grids etc) and launch the kernel (or a series of kernels). Memory transfers are also handled by these wrapper functions exposed through the CUVI API (see cuvi.h header file).

If you can specify your requirement then maybe we’ll be able to help.

Hello Salman UL Haq,

thank you very much for your post. I will take a look at your library. I am new to image processing and I am not familiar with the terminology. The functions I need are the following:

-erosion/dilation

-the equivalent to the watershed function in matlab (to separate two regions)

-filling holes in areas

-calculating the size of an area in pixels

-finding the coordinates of the perimeter of an area

Hello Salman UL Haq,

thank you very much for your post. I will take a look at your library. I am new to image processing and I am not familiar with the terminology. The functions I need are the following:

-erosion/dilation

-the equivalent to the watershed function in matlab (to separate two regions)

-filling holes in areas

-calculating the size of an area in pixels

-finding the coordinates of the perimeter of an area

Hello Salman UL Haq,

thank you very much for your post. I will take a look at your library. I am new to image processing and I am not familiar with the terminology. The functions I need are the following:

-erosion/dilation

-the equivalent to the watershed function in matlab (to separate two regions)

-filling holes in areas

-calculating the size of an area in pixels

-finding the coordinates of the perimeter of an area

hi,

regarding point:
-filling holes in areas

We recently ported a hole-filling (= image inpainting) algorithm to CUDA, it was described in a session of the GPU technology conference 2010 and works quite nice.
You can contact me at ‘hannes.fassold@joanneum.at’ for further details.

best regards,

Hannes

hi,

regarding point:
-filling holes in areas

We recently ported a hole-filling (= image inpainting) algorithm to CUDA, it was described in a session of the GPU technology conference 2010 and works quite nice.
You can contact me at ‘hannes.fassold@joanneum.at’ for further details.

best regards,

Hannes

1 Like

hi,

regarding point:
-filling holes in areas

We recently ported a hole-filling (= image inpainting) algorithm to CUDA, it was described in a session of the GPU technology conference 2010 and works quite nice.
You can contact me at ‘hannes.fassold@joanneum.at’ for further details.

best regards,

Hannes