- ticket title
- iPhone 8 Plus and iPhone 8 now rule DxOMark’s rankings with best smartphone cameras ever tested
- Nest’s cheaper thermostat is better than the original
- Samsung Galaxy Note8 availability expands, arrives in Thailand and Malaysia
- A nine-year-old reviews the Tesla Model X
- The Verge’s TIFF 2017 Awards
Back in November, Google released artificial intelligence experiment that asks you to draw a random object and see if the neural network can identify your doodle. Quick, Draw! was eventually turned into a tool that transformed drawings into clip art based on the best results it got, helping people add a visual icon to their work without requiring any particular artistic talent.
Alongside Google I/O this week, Google has now released the data it received from Quick, Draw! to show you how 15 million people drew the same set of objects. It’s a fascinating look at how humans interpret a random item, from monkeys to parachutes to phones.
Take a look at the Frog one, for example. When I tried this last fall, this is what I came up with.
someone at google never heard of keropi https://t.co/rPyiTyYZDU
— Natt การุณรังษีวงศ์ (@nattgarun) November 17, 2016
I wasn’t very confident of my frog so I felt that adding “Ribbit” to the drawing might add context, even if the AI might not be able to read. It turns out, I wasn’t the only person who did this (though in general, most people didn’t.)
More obscure drawing cues, like yoga, turned out a lot of interesting interpretations.
Face is a neat one too — I’d guess that depending on the artist, most of these drawings were interpreted as a self portrait. Some faces had long hair, some had big eyes, some had glasses, and others just looked like a plain emoji.
The result for dragon is one of my favorites. While you can argue that most people see frog in mostly the same way, dragon produced a variety of results — from fire-breathing to horned.
Google said it has open sourced all the submissions so developers can use the data set for whatever visual-based AI they might be working on.
With thousands of drawings to browse through, it’s pretty easy to lose yourself clicking around the Quick, Draw! datasets. So if you were planning to be productive this Friday, maybe come back to this website after you get your work done.
You must log in to post a comment.