I’d like to use the freespace segmentation GEM with CSI cameras on the NX, and it looks like I’ll need to use the added V4L2 support in Jetpack 4.5 to make that happen.
When I try to use CSI cameras with that GEM now, the camera fails to load properly.
We’re planning on JP 4.5 support in the next Isaac SDK release.
No plans for any Ubuntu 20.04 support yet, however. I personally run everything in containers on Ubuntu 20.04 (focal) host and have no problems, though.
@hemals
given the robot[AGX jetson] has 20.04
so now I want to deploy to the robot from x86_64 Host PC that has 18.04 Ubuntu.
So ther eis a need to create a l4t-base container on the robot, right, so that the deployment happens to the container, as it wouldn’t get onto 20.04 probably, right?
how to customize it so it works like that?
You could in theory build an image based on nvcr.io/nvidia/l4t-base (NVIDIA L4T Base | NVIDIA NGC) and run that container with a shared mount point. You would then copy the runfiles over from your host machine and launch the app from inside the container, following a modified pattern of what is in deploy.sh.