Machine Learning

and

Facial Recognition

Machine learning (ML) is simply a way of making a machine (e.g. a computer) learn without actually being programmed. It is a field where a machine has to learn to do a task by itself. The machine uses data (information), known as “training data” to make a series of predictions or decisions, to learn something. It then uses those decisions to improve the end result - a bit like a human; we are ‘programmed’ to learn from experience. 

Machine learning is closely related to something called “computational statistics”, which focuses on making predictions using computers. Statistics are used to transform the data (information) which you have gathered, into meaningful knowledge.  In the case of machine learning this is done by the computer itself. One benefit is that a computer can transform very large amounts of data and learns as it gathers more data. 

We have created a machine learning model which aims to predict your facial expression! To do this we have used a technology called Google Teachable Machine, a free to use online software which can take pictures & videos through your webcam, or record sounds through your laptop microphone. It then uses them as the training data (as discussed above) to learn which face you are pulling! The more data it gathers, the better it gets at making decisions.

To play with our model, when prompted allow the web browser access to your webcam, scroll down to the display screen and start pulling faces! If you find the model is changing results too rapidly, click the "Enable Frame-by-Frame Classification" and click the "Classify Button"

 

 Some hints and tips for the best results:

  • Really exaggerate your facial expression!

  • Only have 1 person in the camera frame at any time

  • Use the camera against a clear background with your head and shoulders in frame

  • Take off any hats or glasses

You will see that the machine (computer) can decide between 3 facial expressions, but it doesn’t get it right all the time. 

If your webcam does not load, please try refreshing the page.

If you have created your own model, put the URL into our URL box, click "Change Model" and test it out! Make sure the URL ends in '/model.json'

© CLIMB Project, University of Cambridge

Settings

Enable on Frame-by-Frame classification
Classify