Nvidia Jetson¶

First Boot and DNN Image Recognition¶

Need more processing power than a Raspberry Pi, such as for running neural networks for image recognition, or running OpenCV? Check out the following quickstart guide, which walks you through setting up a Jetson on first boot and getting image recognition working.

export:Software files/Jetson/Design Lab Jetson Nano Quickstart Guide.docx

Design Lab Jetson Nano Quickstart Guide Table of Contents¶

• Introduction
• References
• Things You'll Need
• First Boot
• (Optional) Jetson JetPack version / L4T Version
• Hello AI World
  • Image Classification
  • Object Detection
  • Semantic Segmentation
  • Monocular Depth

OpenCV¶

The OpenCV installation you get from apt is one that is compiled without CUDA, meaning that you cannot leverage the GPU for performing graphics-related tasks. There are two ways that you can get OpenCV with CUDA support.

1. Build from source: there are some scripts that can help you build OpenCV with the correct -D flags in order to enable CUDA support. Two scripts have been tested. The first script, https://github.com/mdegans/nano_build_opencv, did not successfully build. The second script by an NVIDIA employee, https://github.com/AastaNV/JEP/blob/master/script/install_opencv4.5.0_Jetson.sh, got further than the first one, but it is not know if it builds fully.
2. Use a container: On this GitHub page, https://github.com/dusty-nv/jetson-containers, one can find many containers that have everything your project needs. OpenCV containers can be found here, where OpenCV with CUDA is built ready for you to use. Make sure that you are using the NVIDIA runtime when running docker run, in the form of docker run --runtime nvidia ....

Robot Operating System¶

Check out the following guide for setting up Robot Operating System for the Jetson!

export