How to build a docker image for arm64 for Jetson Tx2

Hello All,

I am in the process of running a docker image on Jetson TX2. Initially I was getting “Unable to locate package cuda-cudaart-9-2” issue when I tried to build my docker image because of the CUDA dependencies in the image and I could not find a way out to build my image on Jetson TX2. So, I built my image on ubuntu 18.04, an x86_64 based machine and tried to run on Jetson TX2 by importing the built image, but it gave me standard_init_linux.go:211: exec user process caused “exec format error”, which I understood from online means I am trying to run container for other architecture.

I am currently looking to build a docker image for Jetson Tx2, can you please help me in solving this issue.

Thank you.

Hi,

Jeston has its customized OS so you will need to build an image on the top of l4t image.
You can find it here:
https://ngc.nvidia.com/catalog/containers/nvidia:l4t-base

Thanks.

Hi @AastaLLL Thank you so much for the inputs, I built my image on top of l4t image and it was successful.
The current issue I have is, I have many packages to be installed in requirements.txt for the docker image, I see that opencv and scipy are giving me errors when I am trying to build my image.
Do u know how to use the opencv on the base l4t image in the dockerfile and inside the container ? Also, a pointer to install scipy would be helpful.

Thank you.

Hi,

Sorry for the late update.
For OpenCV, you will need to install it from the source in the container.

But we do have a image with scipy installed.
Please check this page for the information: https://ngc.nvidia.com/catalog/containers/nvidia:l4t-ml
This image also have PIL installed. You can try PIL for image usage instead.

Thanks.