Copyright (c) 2021-2024 Antmicro

Antmicro's collection of Yocto layers for machine learning and computer vision applications.

The meta-antmicro Yocto layer consists of the following sublayers:

• meta-antmicro-common
• meta-jetson-alvium
• meta-jetson
• meta-microros
• meta-ml-tegra
• meta-ml
• meta-rdfm-tegra
• meta-rdfm
• meta-antmicro-demo-base

All build dependencies for this project are included in a dedicated Dockerfile. To build a development container image, install Docker and run:

sudo docker build -t yoctobuilder .

This will create a Docker image that can be later used to create the BSP image.

The system-releases directory provides the Yocto build configuration files, as well as the Google repo tool to quickly start testing and development of Yocto-based systems.

For example, to build the system that will run the darknet-imgui-visualization demo with camera feed for one of the Jetson platforms:

• Create the Docker container, which will build the system image in the <build-dir> directory that needs to be specified by the user:

  docker run --rm -v <build-dir>:/data -u $(id -u):$(id -u) -it yoctobuilder
  

  Note: This command runs a Docker container that will be removed upon closing (--rm), mounts the build directory in the /data partition in the container (-v <build-dir>:/data) and builds the system as the oe-builder user ($(id -u):$(id -u)), since Yocto does not allow root builds.
• Configure git settings and fetch the sources using the repo tool present in the Docker container (optionally, you can fetch the sources from your system to avoid git configuration):

  git config --global user.email "[email protected]"
  git config --global user.name "Your Name"
  cd /data
  repo init -u https://github.com/antmicro/meta-antmicro.git -m system-releases/darknet-edgeai-demo/manifest.xml
  repo sync -j`nproc`
  

• Initialize the build environment:

  source sources/poky/oe-init-build-env
  

• Build the system for one of the Jetson targets, e.g. jetson-agx-xavier-devkit:

  PARALLEL_MAKE="-j $(nproc)" BB_NUMBER_THREADS="$(nproc)" MACHINE="jetson-agx-xavier-devkit" bitbake darknet-edgeai-demo
  

  (other available targets are listed in the meta-tegra/conf/machine directory).
• After a successful build, go to the build/tmp/deploy/images/jetson-agx-xavier-devkit directory and untar the built tegraflash package:

  cd build/tmp/deploy/images/jetson-agx-xavier-devkit
  mkdir flash-directory
  cd flash-directory
  tar xzvf ../darknet-edgeai-demo-jetson-agx-xavier-devkit.tegraflash.tar.gz
  

• Put the device in recovery mode.
• Make sure that the device is available for flashing - run lsusb | grep -i nvidia and check if any line appears. There should be something like:

  Bus 001 Device 006: ID 0955:7f21 NVIDIA Corp. APX
  

• Flash the device:

  sudo ./doflash.sh
  

• After the device is successfully flashed, and a camera and screen are connected, you should see the object detection preview.