The Scailable AI manager is the application running on the edge device that allows for easily configurable, and highly performant, edge AI applications. In this section of the docs we explain the full setup and internals of the AI manager.
The diagram below provides a graphical overview of the AI manager as it is running on the edge device. Here we describe each element in turn:
  • The input -- when using our default AI manager installation -- consists of one to eight cameras. Please see our description of suitable cameras and see how to trouble shoot your camera before getting started. Alternative inputs, such as vibration sensors or audio drivers, are available upon request.
  • The AI manager automatically goes through a number of pre-processing steps. Most notably, on each device we support, we provide an extremely low-level and highly-efficient, fully automated, method of grabbing the camera images from the connected cameras and feeding these into the AI model. Through the settings you can control how quickly the images are processed, and you can set areas of the screen to focus on.
  • The runtime executes the AI model that has been deployed to the device in Scailable Portable Model Format (SPMF) using the Scailable platform. Note that the processing to go from the raw input images coming from the camera to the model input tensor happens automatically. Thus, when you are selecting a library model, or when you are creating your own model, you do not have to worry at all about the input camera resolutions. For more details on creating your own models see our documentation for data scientist.
  • The AI model produces raw output which is dictated by the original model specifications and is well-documented for all the available library models. Based on this raw output you can flexibly configure the post-processing which determines how, and when, output is send to an application platform (for example to the Scailable platform, or to your own internal ERP system).
  • Finally, the resulting model Output is send to some configurable location.
Note that the AI manager makes it super easy to simply configure your edge AI solution on the device. You will not need to do any device specific engineering to create and manage scalable edge AI solutions.
In the remainder of these docs you will find information on setting up your AI manager using the UI, and you will find more detailed configuration options in the advanced configuration documentation. Note that everything you see here being done using the UI can obviously also be done in code, using our SDK. Also, every "block" (pre-processing, model runtime, etc.) is easily customizable.
Running an AI model on an edge device using the AI manager often provides output that is not neccesarily visually interesting: for example, the counts of the number of people in front of the camera, or the license plate of a car passing by.
If you are looking to provide interactive demonstrations using the AI manager and would like to, on the edge device, create visualizations of object counts or bounding boxed, please review our on device demo documentation.