Jetson Nano OpenCV 4.1.0 for python3.7

Hi,

we have successfully installed python3.7 and installed openCV 4.1.0 from sources on jetson nano.

We’d like to build a wheel from this installation but didn’t find ant “how to” manuals.
How did jetpack build a wheel for opencv?

The end goal is to have opencv installed without the necessity of all those libraries.

Cheers,
Martin

Hi,

You can try the “Build process” listed in this page:
https://pypi.org/project/opencv-python/

Thanks.

Hi @aastalll,

the instructions there say:

Copy each .pyd/.so file to cv2 folder of this project and generate wheel
Linux and macOS wheels are checked with auditwheel and delocate

that’s it. On jetson nano there is no .pyd file/folder anywhere to be found after opencv installation. The instructions might need some work :) These instructions make sense when building opencv on windows and build linux on docker.

Any reference for instructions specific to arm64 would be appreciated.

Thanks a lot,
Martin

Hi,

Sorry that we don’t have an experience on whl.

Is Debian an option for you? If yes, you can check this page:
https://unix.stackexchange.com/questions/46122/how-can-i-create-a-deb-package-with-my-compiled-opencv-build

Thanks.

nice, thank you for the link

Any news on this?

I’m building opencv4.1 and python3.7 from source as well, because greengrass, which they say is easy to install out of the box, does not support python3.6.

It would be great to have more packages for stuff like this for easing development on this machine.

Hi,

In our latest JetPack4.3 installer, we provide OpenCV4.1.1 with python 3.6.
The python version is chosen by the default python3 in the Ubuntu18.04.

In general, we only provide package built with default python version.
But with some script for other users to build it from source.

Thanks.