Jetson Orin Nano docker error

Built a docker container from the jetson-containers repository by dustynv by combining both the l4t-pytorch and zed container using ./build.sh --name=pytorch_zed l4t-pytorch zed
All tests passed successfully, however when I try to run this code i get the following error.

python3 -c “import pyzed.sl as sl;print(‘ok’);”
Traceback (most recent call last):
File “”, line 1, in
ImportError: libnvbuf_utils.so.1.0.0: cannot open shared object file: No such file or directory

Running this however works.
python3 -c “import pyzed ;print(‘ok’);”
ok

Without using docker I can run pyzed.sl just fine. I am thoroughly confused and wondered if anyone had any ideas on what I can do. Tried looking at /usr/lib/aarch64-linux-gnu/tegra in both the docker container and the Jetson, but both contain the same: libnvbuf_fdmap.so.1.0.0, libnvbufsurface.so, libnvbufsurface.so.1.0.0, libnvbufsurftransform.so, libnvbufsurftransform.so.1.0.0. Strange that it would work on the Jetson when it also doesn’t have the libnvbuf_utils.so.1.0.0 installed.
Running on the Jetson Orin Nano with:

L4T_VERSION 35.4.1
JETPACK_VERSION 5.1.2
CUDA_VERSION 11.4.315
LSB_RELEASE 20.04 (focal)

The entire reason for doing this is I want to run the python custom object detection module in a docker container, which requires both the pytorch for the Jetson and ZedSDK.

Hi @MagnSell,

Welcome to the Stereolabs community :wave:

Can you please share the command you are running the container with? Are you running it with the nvidia runtime as mentioned in this doc: How to Install ZED SDK with Docker on NVIDIA® Jetson - Stereolabs

It may already be the case, but please make sure that the l4t version of your docker image matches the hosts’, as the l4t images mount libraries and files from the host Jetpack.

Hi! I both my jetson and container are running l4t 35.4.1. As for the running command I am running docker run --gpus all -it --runtime nvidia --privileged -e DISPLAY -v /tmp/.X11-unix:/tmp/.X11-unix pytorch_zed:r35.4.1

When I use the official zed:4.1-tools-devel-jetson-jp5.1.2 docker image I am able to run the explorer just fine. The main goal here is to be able to run the custom object detection module so if there is another way to get torch, torchvision etc, into this container i am open to that as well :slight_smile:

Hi @MagnSell,

Probably the simplest way you can create a pytorch + zed docker image would be to create a new Dockerfile based on dustynv’s image, and install the ZED SDK in it. All the ZED SDK tools will also be included.