Some questions about my setup - newbie

Hi. I am a newbie in Jetson. I am working on a Jeson Nano (4gb).
It went like this:
I added the jetson-inference and tensorrt_demos code form github. It all worked.
I installed tensorflow2 using the correct method:
sudo pip3 install --pre --extra-index-url Index of /compute/redist/jp/v45 tensorflow
I realized that there was a major change between tensoeflow2 and 1 - thank you google.
I wanted to use HFNet which is built on tensorflow1 and with all my trials, i could not get the compatibility so I went and installed tensorflow1:
$ sudo pip3 install --pre --extra-index-url Index of /compute/redist/jp/v45 ‘tensorflow<2’
Now the HFNet worked like a charm.
I used mainly opencv for the camera.
I took a look at the code from jetson-interface and saw that they used jetson.utils for the camera.
I tried that and started to have troubles with the camera and the display:
nvbuf_utils: could not get egl display connection – and the like.

I noticed that also the jetson-interface examples were not also broken – with the same comments - although opencv still worked with the camera.

Where did I go wrong? Can I get the jetson.utils with the camera to work again?

Thanks - remember I am a newbie here with the Jetson.


Do you have a physical display connected?
The error indicates that it cannot render the output to a display device.

If you access the Jetson with ssh, please try to set the below environment variable.

$ export DISPLAY=:0



Thanks for your comments - I’ll go a bit more into detail.
In the beginning, I used a dedicated screen for the jetson.
Knowing that the jetson and its attached camera will eventually be attached to a large device, I moved on to using ssh.
I use MobaXTerm to do the ssh for me (actually it should do ssh -X) as the MobaXTerm app has all the stuff I need to support Xterm windows - we work with Microsoft Windows as a primary computer at work so I need to use some external app to get full access to Xterm capabilities.
indeed everything else does open. If I ask for echo $DISPLAY - I indeed get localhost:10. I suddenly recall (while writing this response) from a previous position that somewhere in the /etc/profile that the display was set there - and it was not so good for me. It’s best just to leave it blank. I never checked on the jetson but perhaps this is the issue …

Now when I use the opencv capture stuff for a usb camera, all is well. I can capture and stuff and the application works and I see stuff on my local machine. When I use the jetson.utils to access and capture with the camera - there are 2 statements, one for the camera and the next for the display, I not only get issues with the display but I believe i also see that no images were captured as well.
Since the capture using the jetson is, in my calculation, faster than the standard opencv capture, I’d prefer it.

I’m just wondering whether I ruined something when downgrading from Tensorflow2 to Tensorflow1


Do you still have the access to the dedicated screen?

If yes, could you run the sample on the Jetson directly?
So we can know this issue is from the display or some error when downgrading TensorFlow.



I still (sort of) have access to the dedicated screen. Sort of because the screen is in my office and now the jetson is running stuff on our machine in the production hall lab. I’m waiting to move the jetson back to my room. Alternatively (see my latest post) Im trying to clone the sd card and use another jetson (I’ll change ssh and ip details to get rid of possible clashes) and this can potentially do the check.

Thanks for your continued interest,

There is no update from you for a period, assuming this is not an issue any more.
Hence we are closing this topic. If need further support, please open a new one.

Any result can be shared? Still an issue to support?


This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.