Docker run jetson inference

Autonomous Machines Jetson & Embedded Systems Jetson Nano
#1

Hello,

I am wondering if I can use docker on a pc by changing the architecture from linux / arm to linux / x86_x64?
What are the steps to follow please,

I have already succeeded in completing the project from the sources on the PC, by specifying compute from my gpu card, but for the docker, I don’t know much about it …
do I have to create a new image? or used the one on this link? Docker Hub

thanks

#2

Hello @sylia

In my opinion, the answer should yes. You can refer this dockerfile jetson-inference/Dockerfile at master · dusty-nv/jetson-inference · GitHub.
I’m not sure it will work easily but this may take more time.

Regards

1 Like
#3

The jetson-inference Dockerfile also relies on l4t-pytorch base container, which is for Jetson/aarch64. So you will want to use nvidia/pytorch base container from NGC instead for x86.

1 Like
#4

Thank you for your answers,

the goal of all this is that I intend to launch the training on a cloud (ovh cloud), I wonder if I can directly use the docker image which was generated with the architecture of the ARM64? , because rebuilding in x86 takes time and I have some errors

#5

No, you can’t run aarch64 container on x86 - you would need to rebuild it for x86.

#6

I rebuilt the container for the x86 by modifying the base image NVIDIA NGC,
but i get these errors i dont understand where its coming from?
any suggestions please?error.txt (15.4 KB)

#7

It seems like it might be because they have python3.8 symlink’d to python in that base container.

Try changing these lines to set(PYTHON_BINDING_VERSIONS 3.8)

Also I think you will need TensorRT in the base container. I officially support jetson-inference on Jetson, not x86, so you may find errors you will need to debug.

#8

Thanks, it works

#9

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.