Scripts for creating an L4T base docker image for the NVIDIA Jetson family of devices, with bundled accellerated support for PyTorch, Ultralytics YOLO and dependencies.
The current version of l4t-base supports devices running JetPack 6.2 only.
Notable included packages:
| Package | Version |
|---|---|
| python | 3.10 |
| torch | 2.8.0 |
| ultralytics | 8.3.175 |
Use the build.sh script to automatically build the container. You should run this on Jetson hardware.
Upon running the blank l4t-base container, the self test script will run, followed by the container exiting. NOTE: You MUST use the --runtime nvidia option to enable CUDA support.
alewin@brain:~/git/l4t-base$ docker run --runtime nvidia -it brain/l4t-base:j62-r36.4-1
NOC L4T (Linux 4 Tegra) base image test script
✓ Standard dependencies loaded
✓ PyTorch CUDA support
PyTorch Version: 2.8.0
CUDA Version: 12.6
CUDA Device: Orin
✓ Torchvision loaded
✓ OpenCV loaded
✓ Ultralytics YOLO loaded
If everything comes back without error, you have configured your Jetson correctly!
This is a base container, you should extend from it with your own application.
NOTE: While the supporting build and test scripts are GPLv3, the Dockerfile specifically is MIT licensed.