/jetson_nano_ubuntu20_docker

Hardware accelerated OpenCV, Torch & Tensorrt Ubuntu 20.04 docker images for Jetson Nano containing any python version you need up until the latest 3.12 with ultralytics yolov10 tensorrt support

Primary LanguagePythonMIT LicenseMIT

🐳 Jetson Nano Ubuntu 20.04 Docker Images 🐳

What is this?

Hardware accelerated OpenCV, Deepstream, Torch & Tensorrt Ubuntu 20.04 docker images for Jetson Nano containing any python version you need up until the latest 3.12 with support for ultralytics yolo using torch or tensorrt

Pulling existing images

Python Version Dockerhub Image Name Size
Python3.12 jayfalls/l4t-20.04:full-cp312 Not Ready Yet
Python3.12 jayfalls/l4t-20.04:base-cp312 Not Ready Yet
Python3.11 jayfalls/l4t-20.04:full-cp311 3.28GB
Python3.11 jayfalls/l4t-20.04:base-cp311 1.09GB
Python3.10 jayfalls/l4t-20.04:full-cp310 Not Ready Yet
Python3.10 jayfalls/l4t-20.04:base-cp310 Not Ready Yet

note: Make sure to run the container on the latest L4T host system (r32.7.1). Running on older JetPack releases (e.g. r32.6.1) can cause driver issues, since L4T drivers are passed into the container

Tag Meanings

  • base: This is just the plain ubuntu20.04 image with higher python versions, no additional packages like opencv, etc
  • full: This is the image with all the packages installed, including gpu accelerated opencv, pytorch, tensorrt and deepstream

If an image with your desired python version doesn't exist in the docker hub, or you are running an older jetpack release, then you'll need to build it manually

Building Ubuntu 20.04 Images with Custom Python Versions

Compiled Packages

This link contains all the individual packages for the python versions that get compiled during the build process. If you want to use a different python version without the docker image, you can download the packages and install them manually.

Using Ultralytics Yolo models with Tensorrt

Contributing

You can contribute in the following ways

note: I do not have a lot of free time, so I'm not sure if I'll support this fully, as such don't expect too much from me please. The license is MIT, so you can continue it on your own repo if I don't support.

note: If you see the error ImportError: "/path": cannot allocate memory in static TLS block, you can run export LD_PRELOAD="/path":${LD_PRELOAD} before running your script, then please submit it as an issue so I can put the fix into the build process

References