| Google Uses AI To Make Better Artists |
| Written by Kay Ewbank |
| Saturday, 22 April 2017 |
|
Google has come up with an app called AutoDraw that turns rough drawings into a finished image using a combination of machine learning and a library of pre-drawn images. AutoDraw is available on desktop and mobile devices, and the way it works is that you draw your rough sketch, and the app then works out what it was you actually were attempting to draw. You're then shown a selection of drawings done by more talented artists, and if one appeals to you more than your own effort, you can click on it to replace your drawing with the better one. Some of the suggestions can be a bit harsh - how could they mistake my obvious drawing of a dog for a camel, for crying out loud (though to be fair, there were some dogs among the suggestions too):
According to Google, there are hundreds of drawings that the app will recognise, and more will be added in the future. Google's idea is that this might provide an easier way for non-artists to produce 'professional' drawings than struggling with painting and drawing tools. If the autosuggest mode is too harsh for you, you can turn it off and just use the app as a digital sketchpad instead.You can also add text, insert shapes, and fill areas with colors. When you're happy, you can save it in PNG format. Under the covers, AutoDraw uses the same techniques as Google QuickDraw, which we covered back in November. QuickDraw gives you a target item to draw, then tells you how accurately you've managed it. While QuickDraw aims to improve on a sketch of your choosing, the technology underpinning both is a neural network that can recognise what a bad human 'artist' is attempting to draw. While the documentation about AutoDraw has little background on how the app works, a recent paper published on the Google Research Blog discusses a project to teach machines to draw, and there are many similarities to the way AutoDraw works. In the paper, researcher David Ha discusses a generative recurrent neural network capable of producing sketches of common objects, with the goal of training a machine to draw and generalize abstract concepts in a manner similar to humans. They say: "We train our model on a dataset of hand-drawn sketches, each represented as a sequence of motor actions controlling a pen: which direction to move, when to lift the pen up, and when to stop drawing. In doing so, we created a model that potentially has many applications, from assisting the creative process of an artist, to helping teach students how to draw" |
| Last Updated ( Saturday, 22 April 2017 ) |
