This article describes how to register your Maivin platform with eIQ Portal and how to use the Maivin to capture dataset images, performing validation and profiling, and how models are deployed to the Maivin WebApp Demo.
To get eIQ Portal, follow this link.
Capture Dataset Images
Start eIQ Portal. If you want to create a new project, press the button and choose the name of your project and press "Save".
At the "Data Set Curator" screen, use the "Capture" dropdown button to see if you have any local or remote devices configured. On this PC, we don't.
On the top toolbar of the eIQ Portal application, press the "Remote Devices" button.
On the "Remote Devices" modal, click the "Add Device" button to add the Maivin. We are giving the name "6895 Maivin" and the URI is "verdin-imx8mp-0676895" in our example; your names will be different according to your Maivin's hostname.
Hit the "Add" button to add the Maivin to the remote devices. Now that it is added, mouse-over the device to make the "Select" button appear and select it.
If the device shows up under "Offline Devices", confirm that the Maivin is connected to your network and powered on. After you have confirmed that both of these are correct, hit the "Scan Devices" button to see if it appears in the "Online Devices"; then follow the steps above to select it.
Once the device is successfully added and selected, click the "Close" button to return to the Curator screen. Verify the device is found by mousing over the "Capture" dropdown. If the device does not appear, refresh the workspace by clicking the "Select Model" button to get to the Model Selection screen and click the "Dataset" button to return to the Curator screen.
If you have multiple Maivins configured, make sure the one you want is highlighted yellow.
Now point the Maivin at what you want to take a picture of, and hit the "Capture" button!
Take as many images as you want!
Performing Validation with the Maivin
With larger projects, models trained with eIQ Portal can be validated on the Maivin. Here is a model already trained on eIQ Portal with very high accuracy.
Clicking on the 'Validate' button takes us to the validation window.
Clicking on the "Remote Devices" button on the title bar brings up the "Remote Devices" modal.
Clicking on the "Add Device" button brings up the prompt to add the Maivin. We are going to call this device "Maivin" and use the Maivin's URL of "verdin-imx8mp-06976895" (note: no "http://" or trailing slash).
If successful, the device will show up as an online device.
Clicking on the Maivin and clicking the "Select" button will now allow the Maivin to be used for remote validation.
We should see the Maivin as the remote target on the "Validate" screen once we leave the remote target screen and scroll down the sidebar until the "Validation Stats" dropdown appears.
After clicking the "Validate" button, the "Converting Model" and "Validating Model" will appear. Once they are complete, we'll see the confusion matrix with the results of validation of the model, and the validation statistics can be scrolled to in the sidebar.
Pressing the "Deploy" button will take you to the "Export" page where you can save the model as a Deepview RT model, which we can look at in the next step.
Profiling a model on eIQ Portal
We can profile a trained model using the Model Tool in eIQ Portal.
On the home page of eIQ Portal, click on the Model Tool button.
This brings up the Model Tool window.
The model you exported can be dragged into this screen from File Explorer or loaded via the "File -> Open" menu.
The model is now loaded.
We can use the "Tools -> Manage Targets" window to configure the Maivin. Set the URI to "http://verdin-imx8mp-06976895" and the name to "Maivin".
Hit the "Add" button. You may also want to uncheck the box for the localhost target.
Hit the "x" to close the window, and then on the Model Tool main screen, choose "Tools -> Profile Model". The Model Tool program will grey while the model is uploaded and profiled. You may hit "Cancel" to stop this process at any time.
Once the model is completed profiling, click the "Close" button.
Each layer will have individual inference times. The total inference time can be found in the Model Properties modal by clicking on "View -> Properties...".