I can not link nvidia.box to go on pytorch, how to do?

Robotics & Edge Computing Jetson & Embedded Systems Jetson Xavier NX
1

hello,
when i operate pytorch on NX as nvidia instructions, but can not link to nvidia.box,
how can i do now?

thanks,

3

Hi,

Do you have network connection on your NX?
Thanks.

4

hello，
i worked on NX 5 days and night to try to deloyment our own yolo3 model which trained well.
our project is very urgent to use NV box as a AI platform as edge unit, and i have one few dyas to finish all to show.
i got correct pytorch whl already from a friend in USA and it works now.
but when i load darknet, and modify GPU=1, CUDNN=1, CUDA=1, it can not make. seems it is problem of opencv.
i uses jetpack 4.5.1, and opentv dpkg is 4.1.1.
pls help a link how to fix that?

many thanks!

5

Hi @wilicyy, see here for @mdegans script to build OpenCV with CUDA+cuDNN enabled:

6

Screenshot from 2021-04-07 00-28-44
Screenshot from 2021-04-07 00-28-44735×423 184 KB

hi,
i done your script, and updated cudnn patch, and changed makefile, but it shows much error again…
how can i do now?

thanks

7

After searching it, it seems that darknet doesn’t compile with newer version of OpenCV:

There are some patches for darknet included in those GitHub bug links. Or you can build an older version of OpenCV (I’m not sure which version darknet requires)

8

Hi,

You can find the OpenCV & cuDNN v8.0 patch for pjreddie darknet source below:

Thanks.

9

many thanks for your supports,
and i think opencv is not urgent for my project right now, so i am trying to close opencv and go on next step.
cudnn seems ok.
i will update for new problmes.

thanks a lots!