Nano running Ubuntu 20.04, compiled OpenCV 4.12.0, but python3 doesn’t see it

[JamesNewton]

Copying from here:
Nano running Ubuntu 20.04, compiled OpenCV 4.12.0, by python3 doesn’t see it

Jetson nano on which I’ve installed Ubuntu 20.04 following your install-ubuntu-20.04-on-jetson-nano instructions

The instructions for expanding the available memory don’t seem to work past 4096. Perhaps because it doesn’t have access to the 2048 zram thing. Still… the compile finished

And JTOP shows I have “OpenCV: 4.12.0-compiled CUDA: YES” installed. So… success?

Problems:

python3 says I have OpenCV version 4.7.0 installed (after doing import cv2 and looking at cv2.version) And I DID complete the installation part at the end.

(also a node.js issue, but I won't ask about that here)

Can’t find v4l2ucp because it’s no longer included with Ubuntu 20. But others seem to indicate it's not required, so I just commented that out.

I'm curious how to troubleshoot the Python3 wrong version thing... I see a lot of other people complaining about the same issue and there doesn't seem to be any good advice other than "reinstall" LOL.

I'm looking at nano /home/jetson/.local/lib/python3.8/site-packages/cv2/__init__.py and it's not obvious to me how that works. e.g. where it gets the version information or decides what to pull in.

What I'm saying is this seems like an interesting / worthwhile problem to learn about.

[Qengineering]

I will investigate your issue.
In the mean time, do you know this image with OpenCV ?

[JamesNewton]

I will investigate your issue. In the mean time, do you know this image with OpenCV ?

It is wonderful that you provide these, thank you. It does appear this includes a CUDA enabled 4.8? I don't believe OpenCV 4.8 supports NEON but that is a relatively minor issue, I think.

I will try that image if we can't figure out the Python3 problem, but I would like to first try to learn more about why this is happening. My partition is nearly full, so I'm expanding it and will report back after that is done, and as time allows

[Qengineering]

NEON is supported by OpenCV.
Check:

$ python3
>>> import cv2
>>> print( cv2.getBuildInformation() )

Most likely, your issue is originated by some software package installing OpenCV behind the curtains.
The JetPack you started with has version 4.1.1. So somewhere along the line version 4.7 has slipped in.

[JamesNewton]

Poop. Dead SD Card. Looks like I'm down until I get a replacement card, and my backup is prior to this build, so I'll have to start over when I get a new card.

And it was a SanDisk Extreme, A2, 128GB "V30" whatever that means.

I think instead of top end stuff, the better choice is more than one mid cost card, backups, and one on the ready. Whats a good (not the cheapest, not the most expensive) type of card to buy? I'd like minimum 64GB, and would probably spring for a couple 128s if aren't too precious.

Ticks me off that I'm down right now. I'll be back once I've got my new card, then I will restore, do the build again, and see if I can duplicate the issue again. If not, I'll close this, or you can close it.

[eternaleclipse]

I have followed @Qengineering's post and compiled OpenCV from source a couple hours ago. It is indeed an existing python package that needs to be removed, when you get your setup running, after compilation and sudo ldconfig, if the OpenCV version is still wrong, try:

import cv2
print(cv2.__file__)

Check if the files in there are recent with ls -la (if it's an old package, you can locate and remove it with pip freeze and pip uninstall <package_name>).