Google is trying to develop artificial intelligence that can help users to draw

July 4, 2017, 3:47 p.m. By: Pranjal Kumar


Google is currently working on the Artificial Intelligence technology that can help the users to draw different types of images faster and more accurately. The software that Google is working on is known as Sketch-RNN. It is currently in the form of the web app.

Google’s idea is simple. It will select one of the pre-existing objects and then start drawing the image on the bias of selected image. But, before moving on to actual drawing, the software will try to guess the best way to complete the drawing.

Sketch-RNN is an artificial mind that is trained on a neural network which is fed with many human drawn doodles as the input. It is quite similar to one that was found in Google Brain efforts like AutoDraw and Quick Draw. Google mentioned that the ultimate goal is to train machines to identify and recreate objects with an accuracy of the human mind. The machine should be able to draw and connect lines the way a normal human does while sketching an image of an object.

Sketch-RNN has currently only three demos that one can try “Multiple Predict”, “Interpolation”, “Variational Autoencoder”.

Multiple Predict works like the basic demo. The software will show you multiple possible outcomes at once. Interpolation is a bit tricky compared to the previous one. The system takes two random images and tries to interpolate them to give a hybrid result. The result may be irrelevant as it just shows how to draw a third image based on the other two.

Variational Autoencoder asks users to draw a complete image of something and after finishing up it will try to guess your drawing style and suggest you nine different alternatives based on the way you draw.

Yes, Sketch-RNN might not be as good as AutoDraw but Google is working on it. But, one interesting is that machines are getting way more accurate in visual recognition.

Image Source: