A new AI powered robot called There’s Waldo finds the red-and-white striped adventurer better than anyone.
Put together by Nashville-based creative agency Redpepper, There’s Waldo can spot Waldo on a two page spread with ease.
“While only a prototype, the fastest There’s Waldo has pointed out a match has been 4.45 seconds which is better than most 5 year olds,” reads the video description.
There’s Waldo relies on a robotic arm system to function. The arm is a uArm Swift Pro controlled by a Raspberry Pi. It’s kitted out with a Vision Camera Kit that allows for facial recognition.
The arm extends the camera out over the page. It takes a picture and feeds it through OpenCV to find faces. Then the faces go through Google’s AutoML Vision service.
AutoML launched in January to enable users to train their own AI tools. It doesn’t require any previous coding knowledge. The drag-and-drop tool also makes it easy to build AI.
In the case of There’s Waldo, Matt Reed fed images of Waldo into AutoML. Reed, Redpepper’s creative technologist, lead the project.
He gathered 62 distinct Waldo head and 45 Waldo heads and bodies from Google Image Search.
Despite the small amount of data, AutoML created a strong model and was surprisingly good at spotting Waldos not in the original set.
Some would argue that the machine is cheating, but I disagree. It plays by the same rules we do. We both have to look at the page and recognize Waldo’s face to find him. There’s Waldo is just better — which doesn’t bode well for us humans.
If you’re more interested in the practical applications of AI, Quicksilver is finding scientists missing from Wikipedia and writing them articles.