XNOR.ai develop state-of-the-art on-device AI solutions that enable business to real-time decisions, deliver more efficient experiences to customers.
The following Computer on Modules are supported:
While you can use a wide variety of displays and monitors, additional configuration may be required for a specific setup.
A section in the end of this guide provide instructions about display and touch screen configuration.
This section provides instructions for you to quickly get started with torizonextras/arm64v8-ai2go.
For this demo you will need a camera for image capture and an HDMI screen.
This demonstration requires two containers for its operation. The container with the AI2GO object detection application and the container with a graphic composer, as the AI2GO application will display the captured video with bound boxes to highlight the detected objects:
Another important note is that in order to facilitate the detection model change, AI2GO has several models that can be downloaded, libxnornet.so, detection model file, will have to be stored on the Torizon Core host outside the container. The AI2GO application container will have the model file path shared between the host.
For organization and correct start order of the containers we recommend using Docker Compose:
version: '3' services: weston: image: torizonextras/arm64v8-weston-kiosk volumes: - /dev:/dev - /tmp:/tmp privileged: true ai2go: depends_on: - weston image: torizonextras/arm64v8-ai2go command: /dev/video0 volumes: - /dev:/dev - /tmp:/tmp - /home/torizon/:/xnor privileged: true restart: on-failure
Create or copy the file above with the same content and the name "docker-compose.yaml" on the board running Torizon.
To pull the container images use the below command on the root folder where the docker-compose.yml file is located on the board:
# docker-compose pull
Download any detection model from the AI2GO for Toradex Boards and copy to the your board:
$ scp libxnornet.so torizon@your_ip:/home/torizon
Warning: make sure you copy the model libxnornet.so file to the /home/torizon folder.
Warning: change the "your_ip" on the command above for the IP address from your Apalis iMX8 board.
To run the demo use the below command on the root folder where the docker-compose.yml file is located on the board:
# docker-compose up
You should see the following screen with real-time video captured from the camera with the bound boxes:
In the screenshot above we are using the face detection model.
For more information about the Docker image, Dockerfile, and the source code for this demo see our repo: https://github.com/toradex/ai2go-imx8
Displays and Monitors used in Embedded Systems are available in a myriad of configuration possibilities - resistive, capacitive or without touch, single or multi-touch, different resolution, density of pixels, pin-out and clock frequency are some examples.
To make things easy for you, Toradex provides specific instructions on how to use its display offerings as well as comprehensive information about how to interface your custom display or monitor to Toradex modules.
There are three tested and recommended displays by Toradex:
You can easily set-up and get them running with 3 steps explained on Setting up Recommended Displays with Torizon. If you prefer to configure a specific display, it is suggested the reading about the easy to use Torizon device tree overlays container on Device Tree Overlays.
This release notes are strictly related to the test of the torizonextras/arm64v8-ai2go in Toradex hardware. They are not related to the XNOR.ai releases.