MLADS 2019 - Train an AI model for the Vision AI DevKit

Welcome to MLADS - train an AI model for use with the Vision AI DevKit

What you will do

Setup a Vision AI DevKit camera
Train a model using Customvision.ai and deploy the model to the camera
Train a model using Azure Notebooks and the Azure Machine Learning service
Create an Azure IoT Hub with an IoT Edge device for controlling network traffic between the camera and the cloud

What you will need

Active Azure account
Azure Machine Learning Workspace

Setup the Vision AI DevKit camera

The Vision AI DevKit camera you were provided should be displaying three blinking red LEDs on the front of the camera, indicating the camera is running in Wi-Fi APN mode.
Connect your laptop to the camera’s Wi-Fi APN. The APN for your device can be found on a sticker on the bottom of the camera. The name will be in the format ‘MSIOT_xxxxxx’ (xxxxxx is the last 6 characters of the device’s Wi-Fi mac address, e.g. MSIOT_BD097D). Note: since there are 30 of these devices running in the lab, ensure you select the correct one.
If you have an existing IoT Hub and IoT device for use in this lab, access the Azure portal and retrieve the connection string for the configured IoT device. You will use this string during setup.
Once connected to the camera, use your browser to open setupaicamera.ms. Follow the instructions on the screen and complete the whole OOBE flow. Connecting device to internet is not enough, it has to be associated with IoT Hub as well.
During the setup process, connect the camera to MSFTGUEST Wifi. (No password is required).

Note: After camera setup is complete, you can stream the live camera video output using a browser window or and application such as VLC / Open VC player. Your laptop will need to be connected to the same Wi-Fi network (MSFTGUEST) as the camera to display the video stream.

Train an AI model using CustomVision.ai

You will build a custom AI model to detect when someone is wearing a hard hat, using the Azure Custom Vision service.

Setup up a new Custom Vision project

Login to Azure Custom Vision service at https://customvision.ai.
Create a new project, using these recommended settings:
- Give it a name like Simulated HardHat Detector
- Project Type - [Classification]
- Use the existing resource group
- Classification Type - [Multiclass (Single tag per image)]
- Domain - [General(compact)] NOTE: Ensure you do not select the General option
- Export Capabilites - Vision AI Dev Kit

Upload and tag your training data

Some training images have already been collected for you for the hard hat use case.

Download the .zip file containing the sample images from: https://1drv.ms/u/s!AkzLzaBpSgoMo9hXX4NPjd8QrfhQLA?e=M3ehCL
Uncompress the zip file to a local directory
Upload images to custom vision in batches, one batch per tag (images containing HardHat in the name, then NoHardHat), adding the appropriate tag during upload.

Train your custom model

To train your model using the uploaded training images, go to your Custom Vision project and click on Train.

To export your model, select the Performances tab, then click the Export button. Choose the Vision AI Dev Kit option to download your model. Instead of download press the right button of your mouse in order to see options. Select Copy link. It is a link that points to a zip file of your model.

Clone GitHub

Deploy your custom model to your camera

To deploy your custom model, we will first store your model in a publicly accessible location and then update the configuration of the VisionSample module to use this new model. We will use a cloud blob store to store the model, but any publically accessible file storage system will work.

Upload your trained model zip file

We’ll start by creating a new storage account and then upload your model to it.

Login to http://portal.azure.com.
Search for Storage and select Storage accounts.
Use your existing Azure account and resource group.
Create a new storage account with a unique name (NOTE: upper case characters are not allowed).
Select the West-US 2 region.
Click on Review+Create (other default options should be correct).
Wait until provisioning is complete and navigate to your new storage account.
From the overview tab, click on Blobs.
Click Add a new container, give it a name like hardhatmodel and make sure to select Container (anonymous read access for containers and blobs) for the Public access level.
Click on the container just created, then click the Upload button and select the zip file you downloaded from the Azure Custom Vision service.
Copy the URL of the uploaded file for later use.

Update the configuration of the VisionSample module to use your custom model

Login to http://portal.azure.com and go to the IoT Hub resource you created earlier.
Click the IoT Edge tab, then click on IoT edge device named visionkit.
Click on the AIVisionDevKitGetStartedModule name, then click Module Identity Twin.
Update the zip file with the URL you saved earlier to “ModelZipUrl”: ““, then click Save.

After a few minutes, your device should now be running your custom model!

Test your new model

Put on a hard hat on and smile at the camera!
Verify the camera correctly classifies you as wearing a hard hat.

Train and deploy an AI model using Azure Notebooks and Azure Machine Learning

What you will do

Spin up a VM to your machine learning workspace
Upload data to train the model
Convert the model to a format that runs in Vision AI Dev Kit
Create a container
Create a new deployment.json that specifies the new module for the deployment
Deploy the module to Vision AI Dev Kit

What you will need

Vision AI DevKit camera
Azure Machine Learning workspace

Clone the Azure notebook

Prepare your environment

Change the aml_config/config.json file to match the subscription details for your Azure ML Workspace
In case you want to use your own data go to the data folder and create a folder with your data, e.g. my_data.
- In the my_data folder, copy your data. Each folder name is the label of the images in that folder. For example, the soda_cans folder looks like this:
  - soda_cans
  - coke
  - ice
  - pepsi
IMPORTANT: You MUST have 35 images or more per folder to have enough data for training.