7.1. Edge AI Gallery - User Guide

7.1.1. Overview

The AM62Ax SK board has a QT based Out of Box application setup so that user can test the prepackaged models, with sample images, videos and optionally with a connected camera. The Demo also allows users to add their custom models, images, and videos to extend the functionality.

All of the necessary equipment requirement and instructions are provided here.

7.1.2. Hardware Prerequisites

• TI AM62Ax SK

• PC (Windows or Linux, to use serial terminal console)

• HDMI Monitor (to view the Demo on Display)

• IMX219 Camera

• USB Mouse

• SD card (minimum 16GB)

• Ethernet connection (if serial terminal console is not accessible)

7.1.3. Launching and Using the edge AI gallery

The edge AI gallery launches on Linux startup. Follow the below instructions to use it.

1. Connect the Display, IMX219 Camera, Mouse, Ethernet and Power on the board. The application will launch when the Device is fully booted.

2. Once started, You can click on one of the Image Classification, Object Detection or Semantic Segmentation to validate the models with a fixed GStreamer pipeline.

3. To switch to a new pipeline, click on the same button again to stop the currently running pipeline.

4. To create and run a custom pipeline, click on Custom button. You’ll be provided with a popup menu to choose the below options.

   • Input Type: Image, Video or Camera

   • Input: Image/Video file or Camera Device (For Image or Video files, the path displayed next to the dropdown heading is where you can add your custom files)

   • Model: Models available in the file system (The path displayed next to the dropdown heading is where you can add your custom models)

5. Select the 4 options as described above and click on Start to generate and start the pipeline.

   Note

   As the note suggests, it might take a while for custom pipeline to start.

6. Click on Custom button again to stop the custom pipeline.

7. For ease of use, the IP addr of the device is displayed at the bottom right when Network is connected.

   Note

   The IP addr is polled only for the first 100 seconds after the application has started.

8. To close the application and go to the Wallpaper screen, use the close button on the top right corner.

9. To add new images, videos and models, you should add them in the below directories respectively

   • Images - /opt/edgeai-test-data/images/

   • Videos - /opt/edgeai-test-data/videos/

   • Models - /opt/model_zoo/

10. To change the videos in the prebuilt pipelines used for Image Classification, Object Detection, Semantic Segmentation & Multi Channel, you can replace the video files oob-gui-video*.h264 under /opt/oob-demo-assets/.

11. To relaunch the application, you should run

/etc/init.d/edgeai-launcher.sh start

12. And to stop the application again, you should run

/etc/init.d/edgeai-launcher.sh stop

13. To add new models to the Model dropdown list in Custom popup menu, Add the names of the models in /opt/oob-demo-assets/allowedModels.txt.

Note

For the above changes to take affect, close and relaunch the application.

14. For more information on edge AI software stack, refer Edge AI Documentation

7.2. Compiling EdgeAI GUI App

The ideal way to compile EdgeAI GUI App is to trigger a Yocto build. But for a quicker way to do it, especially during development, TI provides a debian-arm64 Docker image. This image already has all dependencies required for compiling edgeai-gui-app.

First, clone EdgeAI GUI App on host:

git clone https://git.ti.com/git/apps/edgeai-gui-app.git
export EDGEAI_GUI_APP_REPO="$(pwd)/edgeai-gui-app"
cd ${EDGEAI_GUI_APP_REPO}

Then, pull TI’s debian-arm64 Docker image and run it:

docker pull ghcr.io/texasinstruments/debian-arm64:latest
docker run -it -v ${EDGEAI_GUI_APP_REPO}:/root/ti-apps-launcher ghcr.io/texasinstruments/debian-arm64 bash

Finally, run:

cmake -B build -S . -DRT_BUILD=0 # if target is RT image, make -DRT_BUILD=1
make -C build

The compiled binary should be build/edgeai-gui-app.

Copy the compiled binary to the /usr/bin of the target:

scp ${EDGEAI_GUI_APP_REPO}/build/edgeai-gui-app root@<ip-addr-of-device>:/usr/bin/

Note

This is a quick and easy way to compile ti-apps-launcher during development, but it is a good idea to validate with Yocto builds often. There is a possibility that compiler mismatch could present issues in the run up to production.