Tips for understanding CUDA code and its usage with common libraries

[palatinusmagnus]

Hello,
My goal is to utilize the CUDA gpu features of the Jetson Nano to do robust realtime video processing.
Are there any guides, tips, or resources for understanding how to actually USE cuda features of the Jetson?
I dont want to know how to install it. Just the concepts of the code from a fundamental level and how its meant to be used in a similar application. Ive seen videos, but for the past week cannot tell if its my compiler that spits out errors or my code that spits out errors because I have no baseline.

I have an end idea in mind, and a python way to do it, but I just cant think of the right search terms for the right technology of what I need to grasp to complete my task.

I just want to do blob tracking at 60 fps from a 1080p CSI camera to be honest.

Theres all this DeepStream SDK, FastVideo SDK, and all sorts of things.
Im trying hard to try to convert opencv Python code to C code, then from C code to Cuda code, but its sort of a black box mess going on.

Any tips or resources would be greatly appreciated.
Thanks
Yelizar

[AastaLLL]

Hi,

For video processing, we recommend users try our Deepstream SDK since it has optimized for Jetson.
Do you want to implement the CUDA code on your own?

If you want to use a vision-based algorithm with OpenCV, maybe you can check the GPU implementation first:
https://docs.opencv.org/4.x/d1/d1e/group__cuda.html

Thanks.