On device demos
The Edge AI Manager includes a basic visualisation engine. It contains a universal counter that works with many different Scailable models and an emotion visualisation that works for the Scailable emotion recognition model.
You can view real-time results of the AI detection with the visualisation engine.
You need to select a model that is supported by the visualisation, set in and output correctly, and the inference needs to run in order to be able to display the visualisation.
The universal counter displays bounding boxes and counts certain classes of objects. The default engine contains object localisation, line crossing, object recognition, ANPR, bar code reading, alarms, a universal counter and an emotion recognition visualisation.
The types of models that can be used are models are automatically recognized in the AI manager, and when running a button to access the visualisation will automatically appear. Some examples that work are:
- Car location model
- Face locator
- People and vehicle alarm
Start the inference engine as usual, by clicking the "Run" button on the "Run" tab
Accessing the visualisation can be done in a web browser by clicking the "View live visualisation" button in the Edge AI Manager interface.
Alternatively the demo is available at the same location as the Edge AI Manager but with a
/demo/appended to the URL. So if your Edge AI manager is accessible at
http://localhost:8081/the default visualisation will be available at
The visualisation will show the latest image and depending on the model that is used a counter and overlayed bounding boxes. It will update a few times a second, depending on the model.
The visualisation needs a short time to get started, usually a second depending on your computer.
The model might be incompatible or the input might be unusable for the model.
If you test the model in the AI manager the output should show a part with
bboxesin the json output. If the bounding boxes part shows all zeroes the input is not recognized as on of the classes. If there is not bounding boxes output you might try another model from your library.