How to use cv::Mat with NPP

Accelerated Computing GPU-Accelerated Libraries
1

Hi all,

I am currently programming a computer vision program using openCV and cuda. To accelerate the speed of the watershed algorithm I want to use the NPP library. The only problem is that I can’t seem to get the cv::Mat (or GpuMat) to work with NPP.

I tried copying the data from the Mat to Npp8u (which is REALLY slow). Also tried to create a Npp8u pointer of the Mat, this just would not work.
Also in all the examples of NPP i found that they only use .raw images that are stored on the HDD/SSD.

So I have a few questions:

  • Can I use anything other than .raw images?
  • Must the image be stored on the HDD/SSD? (so no live camera feed)?
  • Can I transform my RGB cv::Mat to be usable with NPP functions?

some info:

  • OpenCV version 4.5.3
  • Cuda version 11.6
  • OS: Ubuntu 20.04

I hope somebody knows the answer to all, or one of, my questions.

2

Okay, call it typical. After searching the whole day I find nothing.
But when I ask this question here, and continue searching, I find it almost immediately…

For anybody with the same question, look at the following link, it explains how to use Mat and GpuMat with NPP.
https://titanwolf.org/Network/Articles/Article?AID=f80d409d-2a38-4c55-9236-364a5ee42651

3

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.