Innovation Hub
An initiative by Al Bayat Mitwahid in UAE
Course 2
Using and Training AI
Unit 3 – Creating your own training data
This unit will focus on how to use your own training data with an AI program. Most of the common generative AI programs are useful because they have been trained on datasets that the creators knew would create useful results, for example coding examples or a wide variety of images. It is sometimes useful to create a more personalised AI system that uses your own data, so that you can use the program for your own specific use. This is where your own training can supplement the training of an AI.
Lesson 1: Google teachable machine
Google has created a teachable machine resource that you can use to create an AI program to train with your own media data. You can choose to teach it on image data, sound data or data about the position of a human. We will train it on image data from the webcam.
​
INTERACTIVE
Go to https://teachablemachine.withgoogle.com/ Click Get Started, and choose New Image Project, then Standard Image Model.
​
You will see two grey boxes on the left called Class1 and Class2. We will add images of a different subject to each one, and then the program will train itself on those images.
When we run the trained program, it will be able to tell one type of subject from the other.
Find two different objects that are similar - for example:
a cup and a bowl,
a book and a tablet or
an apple and an orange.
You could even use two different people!
​
-
Click on the webcam button for Class1, and position the first object on the screen. Hold the record button and move the object around until it has taken at least 30 images of the object. You should make sure it has captured the front, back, sides, top and bottom of the object.
-
Do the same thing for Class2 using the other object.
-
Click the Train Model object, and the program will analyse the images to generate a model to “understand” each object.
-
Change the names Class1 and Class2 so that they show the names of the objects.
-
When it is ready, test the model by placing one of the objects in front of the webcam. The program should be able to recognise each object.
-
You can add more classes so that the program can recognise many different types of objects.
If you have time, you can try out the sound or position projects by starting a New Project.
