Google displayed a few of its funky machine learning experiments in November, and one of them was Quick, Draw!. It is a game where you sketch something and an image recognition system guesses what it is. Now the company is releasing the millions of sketches players submitted as an open data set for AI developers to play with.
Now, if an idea of browsing through a collection of terrible drawings seems boring to you, don’t worry. That’s not really the case.
Actually, it’s all about the metadata. Those drawings came from different parts of the world, and it’s actually fun to check out how creatively, say, the UK and Korea think of cats. Or chairs!
Well, Korean and Russian were more likely to draw the chair at an angle or sideways. Why? That’s what you have to find out with your machine learning system!
There are of fascinating differences. As Google’s post calls attention to, an inclination showed up toward the sneaker-type shoe, to such an extent that the system would experience difficulty perceiving a high heel or sandal.
Google proposes you utilize their new Facets tool to visualize the colossal amount of data in the set. What’s more, that is truly the fascinating part of this entire thing. When you have data sets this tremendous — in a huge number of examples— how might you sort it and watch it even at a gross level so individuals can discover coarse examples and thoughts worth seeking after? Or how might you spot things like systematic biases or opportunities for development?
The 50 million drawings are only a beginning — the other 750 million or so will be released after some time, and probably intriguing data from different projects also.