TensorFlow is an end-to-end open source platform for machine learning. https://www.tensorflow.org/

In this workshop, we will use supervised learning to build a model that can be used for image classification. TensorFlow supports a variety of different input types, we will be using images to build our model. After creating a model, we will be able to pass in data and get results with confidence levels.

NOTE: Machine Learning is a big topic. We will stay high level with this workshop by using tools provided by TensorFlow for quickly creating models that we can use with TensorFlow.js.

There are many different approaches to machine learning. We will be using supervised learning which is the process of providing input data and classifying that data into predefined buckets. This approach is good when the input and output parameters are well defined (e.g. wearing a facemask or not wearing a facemask). Depending on the problem you are trying to solve, you may want to explore other approaches.

Results from supervised learning models will depend greatly on the quality and quantity of input provided when creating the model. A small sample dataset will result in poor classification. It's important to thouroughly test your model before using it to make decisions in a real world setting. Keep in mind that Machine Learning models can and often do still have bias.

If you've done any web development in the past, you likely have most of these setup already.

• Node.js
• GitHub account (optional but recommended)
• git command line tools (needed for deployment)
• Heroku account (needed for deployment)

There is a sample model provided that has been trained to detect a face mask. This model was trained just a handful of images and will not be very accurate. After running the test, try training a model of your own!

After downloading or forking and cloning this project. Open the project folder using terminal and type the following command:

npm install
npm start

Visit http://localhost:5000 in your browser and click to allow access to your webcamera.

NOTE: Some browsers may require https in order to access the webcamera.

• https://teachablemachine.withgoogle.com/train/image
• Train your model
• Test your model using your web camera
• Export your model
• Unzip your exported model and copy the containing folder into the public folder of this project
• In client.js, update const URL = './demo-image-model/'; to point to your model
• Stop your local server by pressing ctrl and c at the same time in your terminal window
• Type npm start to restart your server
• If you're familiar with HTML / CSS, try adding some styling to the page.

• Did you provide enough input images?
• Do the input images represent users of your application?
• What gaps might exist in your model?
• Are you able to 'trick' your trained model?

Once everything is working the way you would like, you can deploy the project to Heroku.

• Sign up for a free account on Heroku
• Stop your local server by pressing ctrl and c at the same time in your terminal window
• Make sure your project is a git repository
• Type heroku create in terminal
• Type git push heroku master in terminal

If you get the error Error: The implicit shape can't be a fractional number. with the image classifier, try a different image. Using screen shot images do not appear to work.

• https://www.tensorflow.org/js/demos
• https://becominghuman.ai/image-classification-machine-learning-in-node-js-with-tensorflow-js-dd8e20ba5024