Docker images built for the wrong architecture

I’m trying to run a docker container from ngc.nvidia.com with the suggested command:

sudo docker run -it --rm --net=host --runtime nvidia  -e DISPLAY=$DISPLAY -w /opt/nvidia/deepstream/deepstream-5.0 -v /tmp/.X11-unix/:/tmp/.X11-unix nvcr.io/nvidia/deepstream-l4t:5.0.1-20.09-samples

but multiple missing shared objects are reported by Gstreamer:

(gst-plugin-scanner:14): GStreamer-WARNING **: 17:09:01.443: Failed to load plugin '/usr/lib/aarch64-linux-gnu/gstreamer-1.0/deepstream/libnvdsgst_infer.so': libnvbufsurface.so.1.0.0: cannot open shared object file: No such file or directory

and the target ahitecture seems to be wrong…

WARNING: The requested image's platform (linux/amd64) does not match the detected host platform (linux/arm64/v8) and no specific platform was requested

Upon inspecting the docker container with

docker inspect nvcr.io/nvidia/deepstream-l4t:5.0.1-20.09-samples

the architecture is indeed listed as amd64:
"Architecture": "amd64"

When building my own image I managed to solve the missing libraries issue by copying libraries from the jetpack image file, but what can I do about the wrong architecture? Thank you!

After a lot of research and debugging, I found a similar thread from Pi-hole:

Apparently, in Docker 19.03 the target architecture was ignored, so everything worked fine, but the behaviour changed in Docker 20.03. In the end I have just reflashed my jetson nano and the warning disappeared. Still, the missing libraries problem persists.

It would be great if Nvidia could provide some official info over this.

Are there any news about this?