Published On: Mon, Aug 28th, 2017

Google releases millions of bad drawings for we (and your AI) to duke through


Back in November, Google showcased a few of a musty appurtenance training experiments, and among them was Quick, Draw! (their bang, not mine) — a diversion where we blueprint something and an picture approval complement guesses what it is. Now a association is releasing a millions on millions of sketches players submitted as an open information set for AI developers to play with.

Now, if a awaiting of browsing by a bunch (I’m articulate 50 million here) of terrible drawings of hats, shoes, and cats doesn’t sound like fun to you, don’t worry. That’s not unequivocally a point.

Really it’s about a metadata. Those drawings came from lots of opposite countries, and it’s fun to check out how differently, say, Germany and Korea consider of cats. Or chairs!

Oh.

Well look, there are patterns in there value sussing out. Clearly Korean and Russian sketchers were some-more expected to pull a chair during an angle or sideways. Why? That’s what you’re ostensible to find out with your appurtenance training system!

There are indeed lots of engaging differences. As Google’s post points out, a disposition seemed toward a sneaker-type shoe, so most so that a complement would have difficulty noticing a high heel or sandal. And what about cats? Surely there are sub-cat-egories of how people chose to pull them. we consider we did one, actually, and we drew a whole cat. Am we some strange, large exception? If usually we knew how to emanate a appurtenance training complement to find out (hint).

Google suggests we use their new Facets apparatus to daydream a huge volume of information in a set. And that’s unequivocally a engaging partial of this whole thing. When we have information sets this large — in a hundreds of millions of examples — how can we arrange it and observe it even during a sum turn so that people can find counterfeit patterns and ideas value pursuing? And how can we mark things like systematic biases or opportunities for improvement?

The 50 million drawings are only a start — a other 750 million or so will be expelled over time, and presumably engaging information from other projects as well. Keep an eye on a Google Research blog (or TechCrunch, obviously) for a latest.

Featured Image: Google Research

About the Author

Leave a comment

XHTML: You can use these html tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>