Jump to content

Program Beats Humans at Recognizing Sketches


Recommended Posts

Personally I am not afraid that a robot-apocalypse will happen, but when I see news like this, I cannot help but feel we should be working harder to maintain our dominance. A recent study from Queen Mary, University of London has determined that the program Sketch-a-Net is better at identifying sketches than people. The program had a success rate of 74.9%, beating humans at our 73.1% rate.

Sketch-a-Net is a deep neural network, which means the program emulates how our brains process information, but its success comes from more than that. The program uses information that is normally discarded, including the order the strokes were made in, and it turns out that information is helpful for identifying the subject. The program does need all help it can get too, due to the abstract nature of sketches, compared to photographs. Still it was able to distinguish the finer details of some sketches, including identifying four different bird variants.

With the growing use of touchscreens, Sketch-a-Net and similar programs will become ever more important as a means to interface with our devices. It could also find more professional uses, such as in police forensics for identifying persons in mugshots or CCTV footage from artists' impressions, as well as image retrieval systems.

Source: Queen Mary, University of London

Back to original news post

Share this post

Link to post
Share on other sites


  • Create New...