Comment on page
Getting started
Here we provide a quick "getting started" guide demonstrating how to configure the AI manager (by logging into the edge device), such that your edge device takes a camera input and, on a second-by-second basis, submits a count of the number of detected people in front of the camera and logs the resulting counts to the Scailable Platform.
This getting started guide is a "get started quick guide". That means we are assuming you have a device available, the AI manager installed, a camera available, and all that. The guide contains links to more elaborate documentation in case you are missing out on any of these steps.
For this quick getting started guide we assume that you:
- 1.Own one of our supported edge devices with our on-device AI manager installed. We also assume you have the device that runs the AI manager connected to a network such that you can reach it remotely and such that the AI manager can establish a connection to the Scailable platform.
- 2.Have access, within a network accessible by the device that runs the Scailable AI manager, to a camera stream (or static image) that you want to test on.
- 3.Have a Scailable platform account, potentially a trial account, to retrieve a license for your AI manager and couple your edge device to your Scailable platform account.
The Scailable on-device AI manager (enabling your Scailable pipeline) will run on virtually any Unix device. Please see our generic AI manager installation docs, or our device specific installation instructions.
If you are in a hurry to get started, open your device shell and try running:
sudo bash -ic "$(wget -q -O - https://get.sclbl.net)"
Ensure that you have (remote) access to your edge device that has the AI manager installed, and navigate to port
:8081 on the device. Using any webbrowser you will see the Scailable AI manager interface as it is rendered on the local device:
Note that at this point it is useful to give your Device some meaningful name by simply typing in your desired name (I will call it "Scailable docs demo") and see if all the Status checks are succesful. We should have:
- 1.Connection successfull. There is a connection to our cloud.
- 2.Module ready. The configuration part of the AI manager is properly running on your device.
- 3.Deamon ready. The AI inference part of the AI manager is properly running on your device.
- 4.Output distributor ready. The output distributor, a small bit of software that allows your to both locally and remotely log your output data, is running.
If any of the above statusses is not ok, please check our AI manager status page to see what the status messages mean and how you might be able to solve the identified issues.
If all the checks are ok, your device installation is all set. However, you might still need to register your AI manager license. The steps are easy:
- 1.Click the "Register" button on the AI manager.
- 2.You will be taken to the Scailable Platform (see figure below), where you can either sign in using an existing Scailable platform user account, or you can create a new account.
- 3.After signing in, we will check whether your device already has an associated license. If so you can simply claim the device and go back to the AI manager configuration. If not you will have to provide some details to retreive your device license. The steps should be self explantory. After filling out the neccesary details to retrieve a trial license, you can go back to the device by clicking the "Back to Device" button.
You will return back to the AI manager setup menu on the device (again, running at port
:8081), but this time you will have a license key assigned.
You are now ready to go and configure the AI manager on your device. First, let's select an AI model under the "Model" tab:
By clicking the "Select Model" button you will be taken to the Scailable platform where you can see a list of all the models available for your device. Here we select the "People counter" model and return back to the AI manager on the device.
With the AI model selected, it's time to select the model's input. For demonstration purposes we select one of the example images included in the AI manager. However, you can fill out the URL of your camera stream in the "Input URL" field and select the appropriate encoding, after which you will see a preview of the input image. If you don't see a preview, please see the AI manager input configuration docs.
To select an example image switch to the "Example sources" tab and click on one of the buttons, e.g. "faces" because the model will attempt to locate people.
There are many more input settings you can choose, you can mask the camera input and get access to secure cameras. For more information and details settings see the AI manager input configuration docs.
We now have an AI model selected (the People counter), and we have selected an input image. Next up is specifying the output scheme and location of the resulting inferences (i.e., the people counts). Here we set the AI manager to post the resulting output to the Scailable data distribtor. This will send the data from your device to the Scailable platform such that you can view the results. This setting is our default setting.
You can do way more with the output of the AI manager; you can flexibly configure alarms, and you can send the data to your own desired application platform. You can even capture images for model (re-)training. You can find more information in our AI manager output configuration docs.
For this getting started guide we skip over the "Test" tab which is documented in detail in AI manager docs.
Now that you have selected a model, the input, and the output schedule, you can set the device to work by going to the "Run" tab, and subsequently clicking the "Run Model" button. After clicking, it looks like this:
Which means that now your model is running, its generating inferences, and the resulting counts are send to the Scailable platform. Clicking the "View output in cloud" button get's you to the data logged by this device:
That's all there is to it.
Depending on your needs, your next steps might be:
Last modified 19d ago