How can I enable GPU for real time person detector?


Please forgive my ignorance I am an absolute newbie, probably way over my head.
I am trying to build a fairly accurate person detector with the Jetson Nano (4G) using OpenCV with Python3 and Yolov3-tiny (live feed from 8MP Camera Module IMX219 Infrared).

I have it working but its far too slow for real time (I would hope for at least 15 frames per sec). I understand that one option is to employ the GPU, I have searched for an easy way to do this but the more I look the more complex it seems to be.
I would be very grateful if someone could put me on the right path?

What I understand is that I will have to reinstall OpenCV with Cuda enabled. Then I would have to, in the code, use an OpenCV function to assign tasks to the GPU then pipe the output back to the main body (CPU).
Sounds like building a space ship to mars may be easier :-)

Any link or advice would be enormously appreciated.


Hi @neilvs,

Looks like your query is not related to Tensorrt. This forum talks about updates and issues related to Tensorrt.
Please post your query in related forum.

Thank you.

sorry. Will do

1 Like