OpenCV GPU acceleration and profiling

Hi there,

I’m currently developing some computer vision software on the Jetson TK1. I have installed OpenCV on the Tegra via the Jetpack package (running L4T 21.4). I have compiled and run my software on the Tegra, but didn’t notice a significant increase in speed from using the GPU.

To make use of the GPU do the OpenCV GPU functions need to be used, or can regular OpenCV functions be used? I ask because the elinux wiki implies that you can write regular OpenCV code, which will automatically be accelerated.

I am also looking to profile the performance of the CPU & GPU while running my software. I have used gprof to check out the CPU, but as far as I’m aware gprof doesn’t profile the GPU. Is there a tool for doing this?



Regular OpenCV code is not automatically accelerated. You need to use the GPU accelerated OpenCV functions.

Yes there are various CUDA profilers which are covered in the documentation.