dusty-nv / jetson-inference

Hello AI World guide to deploying deep-learning inference networks and deep vision primitives with TensorRT and NVIDIA Jetson.
https://developer.nvidia.com/embedded/twodaystoademo
MIT License
7.74k stars 2.97k forks source link

Jetpack 5 vs Jetpack 6 Missing Files #1865

Open ahuckphin opened 3 months ago

ahuckphin commented 3 months ago

I have no issues with running jetson-inference on my Orin Nano running Jetpack 5. However when running Jetpack 6, following the exact steps I used previously, I ran into issue of missing files after docker/run.sh.

Jetpack 5 screenshot: jetpack 5 screenshot

Jetpack 6 screenshot: jetpack 6 screenshot

I am able to restore the files per se by running the build commands. However, I am wondering if this difference is intentional and/or there's a better workaround. Cause even after building the files, experience on Jetpack 6 continues to be different than on Jetpack 5. For example, a usable image directory containing the sample images is in a different location.

I also noticed that on Jetpack 5, after docker/run.sh, I am automatically in a jetson-inference directory. However on Jetpack 6, after docker/run.sh, I am in root directory and need to first change directory (cd) to a jetson-inference directory.

bluevisor commented 3 months ago

docker/build.sh asks me to specify a tag, which tag should I use for L4T R36.3.0?

JoostdeK commented 3 months ago

Seems like same issue as #1860

dusty-nv commented 3 months ago

Ahh, okay sorry everyone - on JetPack 6, I have moved jetson-inference docker over to being built in jetson-containers (this is so that will automatically be built on top of the rest of the stack instead of relying on static base images, and is able to be used when building other containers on top)

The idea was still to be able to use jetson-inference's docker/run.sh script to run it - but there was that issue with the symlinks, which should now be fixed in these commits:

I rebuilt and tested it again here, and the contents are now showing up properly under /jetson-inference. When you start the container with docker/run.sh on JP6, it will show the working directory as /opt/jetson-inference, but these are linked (by convention, in jetson-containers I build packages under /opt as to not pollute the root directory)

I also confirmed you can run things like detectnet.py images/peds_0.jpg images/test/peds_0.jpg, and when the container was started with docker/run.sh you will see the images show up on your host under jetson-inference/data/images

I have re-built and pushed the updated images for JP6 to dustynv/jetson-inference:r36.2.0 and dustynv/jetson-inference:r36.3.0, so do a docker pull on those and let me know if you still face the issue.

ahuckphin commented 2 months ago

Issue appears to be resolved now. Thanks.