Microsoft Sponsored Hackathon At Oxford University Mathematical InstituteUsing Microsoft's Cognitive API to introduce Oxford University students to Face recognition, Optical Character Recognition and Facial Comparison.
Playing Who’s Who with real time image analysis
Using Microsoft’s AI Services for image analysis
to produce the AI Gaming version of Who’s Who.
Analyse all of the images you are given in order to detect all of the faces and any text in the images.
Choose a suspect for your opponent to identify.
Identify your suspect by asking questions to eliminate people.
“Is the suspect wearing a hat?”
“Does the suspect have blonde hair?”
“Is the suspect smiling?”
Once you know who your suspect is, you must then identify them in a line up that includes similar looking people.
Compare the face of your suspect with all of the faces in the line up.
How To Play Our Who’s Who Game
The Initial Images
Presented with a grid of faces, the challenge was to use Microsoft’s Face API to identify each face and to determine as many attributes about each face as possible.
Hair colour, glasses, headwear, age, gender and emotion are just some of the attributes that people took from each image.
Additionally, the person’s name was part of the image and this had to be read using Microsoft’s Computer Vision API’s optical character recognition.
With all of the faces analysed, it is time to eliminate suspects by asking questions to narrow the list of suspects. “Is the suspect wearing reading glasses?”, “Is the suspect male?”, “Is the suspect Smiling?”, and finally, when you have your suspect, “Is the suspect’s name Aaron?”
Identifying The Suspect in a Line Up
With your suspect correctly identified, there is still one more step. You must identify them in a line up of similar looking people by identify the alternative photo of your suspect.
Developing The Solution
One of the key elements of the AI Gaming platform is that you can play a game as soon as you register. We give you the template code that is enough to play a complete game from start to finish… It just doesn’t play a very good game.
Working template code means that you can see exactly what is required for each stage of the game. Instead of having to concentrate on length instructions and requirements, you can start to compete in the hackathon straight away and every tweak you make to the template code improves your chances to win in the next tournament.
To test people’s code and to play for prizes at the event, everyone’s bots were pitted against each other in several tournaments.
Three tournaments were played throughout the evening with the winner of our first tournament at 7:30pm winning a Raspberry Pi.
The next two tournaments were combined to determine the overall winner for the evening.
The Results – Our Competition Winners
Our first place winner walked away with a Microsoft Internet Of Things kit and an HD Drone
Only two points behind the first place winner, second place took a Microsoft Internet Of Things Kit.
Third place took a slow and steady path the success. Consistent play earned a Raspberry Pi Starter Kit.