OpenCV, Tensorflow cuda support in jetpack and container

Robotics & Edge Computing Jetson Systems Jetson Nano
1

Hello All,
I’m using the latest jetpack 4.5 on my nano and would like to use OpenCV and Tensorflow on my board with CUDA support.

  1. The jtop command indicates that opencv 4.1.1 is installed but cuda support is not enabled. Why is the jetpack offered without CUDA support?
    How can I enable CUDA support for open CV?

  2. The container here [https://ngc.nvidia.com/catalog/container/nvidia:l4t-ml] has both packages that I need. Does this container have tensorflow and opencv compiled with Cuda support?

Thanks for the response.

Kind regards,
Kashyap

2

Hi @dusty_nv @AastaLLL
I’m a bit lost as to how I can get starting with opencv with jetson inference (docker/without docker).
Can you please answer above questions and this post?

Thanks,
Kashyap

3

You can try this script:

4

Hi @dkreutz Thanks for chipping in. This time I used a mix of old nvidia script by modifying it for other installations.
https://raw.githubusercontent.com/AastaNV/JEP/master/script/install_opencv4.5.0_Jetson.sh

Thanks!

5

My question is this.

7

Hi,

  1. You can enable it by building OpenCV from source.
    Below is a auto-build script for your reference:
    JEP/install_opencv4.5.0_Jetson.sh at master · AastaNV/JEP · GitHub

  2. The container has TensorFlow and default OpenCV.
    You can enable the OpenCV with CUDA support with the script shared above.

Thanks.

8

Hi @AastaLLL
Thanks for the response, will try.

Thanks,
Kashyap