Sadly, not everyone can say they have an artistic flair. It’s predicted that around 10,000 hours of practice are needed to be able to become proficient with a pencil and paper and most of us don’t have the time (or the patience).
Thankfully, Google is taking its artificial intelligence systems and applying them to artworks to make drawing easier for everyone. The tech giant has released a free web-app that allows for terrible digital drawings to be turned into recognisable objects.
Called AutoDraw, the system has been dubbed as “fast drawing for everyone”. It works by providing a blank canvas, not dissimilar to Microsoft Paint, and allowing for line drawings to be made using a mouse or finger if a touchscreen is being used.
As well as simple lines, text, a paint bucket for filling, different colours and pre-determined shapes can be added to the canvas. When creating a masterpiece using the Auto Draw brush, Google’s system attempts to identify what is being depicted and offers a better version of it.
For example, WIRED’s terrible attempt to draw a cow (above) resulted in being provided with the option of bears, dogs, big cats and a rhino (which we settled with as a suitable alternative). It’s then possible to click on the suggestion and be provided with a professionally drawn version. The resulting image can be downloaded, shared, or used at your leisure.
A second attempt – to draw a non-descript smartphone – resulted in the system correctly working out what was being badly created and offering a better version.
Google says the system works by comparing what is being drawn with pictures that have been created by a selection of artists it has worked with. The firm’s machine learning technology has been trained on a dataset of images and is able to compare them to what is being drawn by the user.
AutoDraw is the latest AI-based experiment Google has released into the wild. In November 2016, Google released a series of machine learning games, designed to help train its systems to be smarter and gather data for them to learn from.
Within the games was a Pictionary-style creation called Quick Draw. The game presented users with a named object and asked them to draw it the best they could in a short amount of time. Quick Draw would then try to correctly guess what had been drawn.
At the time the game was released, Google had published eight separate machine learning experiments. Elsewhere within the collection is a AI keyboard that you can play a duet with, an example of what a neural network can see and an object-based language translator.