Computers have come long way since the days of their inception. They can now play chess and beat human opponents, perform humongous calculations and do a lot more. To add to the list, researchers at Brown University have now created a computer program which can identify abstract sketches.
If you make an accurate sketch, it is easy to discern for a computer. For instance, if you draw a cat just like a real cat is, a computer can match a few features and tell you that it is indeed a cat. However, if you intend to draw a cat, end up creating somewhat of a cross between a cat and a rabbit that is hard for a computer.
A human reader may be able to identify it as a cat due to his own understanding but computers don’t work like that. Or at least they didn’t, until now. Brown University researchers say that they have created the first computer app which can accomplish ‘semantic understanding’ of abstract sketches.
This means that even if you actually wanted to draw a rabbit and ended up drawing the Bunny from Bugs Bunny, the computer will be able to identify it. In other words, the new computer program is able to put an abstract sketch in context to the other details which may be related to it – and based on that knowledge, it is able to identify what exactly has been drawn.
Currently, the database of the app holds 250 categories of objects, each of which holds thousands of objects. The app is able to correctly identify a given object 56 percent of the time, which is impressive especially when considering the fact that humans get abstract sketches right only 73 percent of the time.
Interestingly, the computer program doesn’t make use of any particularly clever algorithm. Rather, it simply has a fairly large and intelligently developed data set which it uses to identify the objects.
According to the assistance professor of computer science at Brown University, James Hays, “It was the data gathering that had been holding this back, not the digital representation or the machine learning; those have been around for a decade. There’s just no way to learn to recognize say, sketches of lions, based on just a clever algorithm. The algorithm really needs to see close to 100 instances of how people draw lions, and then it becomes possible to tell lions from potted plants.”
Below is a video demonstrating how the computer program works:
Source: Brown University