While we struggle to supplement our common sense, there is a computer program running 24 hours a day building up its common sense in image processing. The program called Never Ending Image Learner (NEIL) is continuously searching the Web for images and is trying to understand them.
The program NEIL understands those images. After processing, it identifies and labels the parts of images too. It also recognizes the objects in the image. i.e it can identify the rockets and A.K.47 if it encounters them.
Besides understanding the visual world better, NEIL also makes associations between things that we humans know. For example, it can obtain the common sense information like cars are found on the roads, car can have a part called "wheel" and much more.
"Images are the best way to learn visual properties," said Abhinav Gupta, assistant research professor at Carnegie Mellon University's Robotics Institute. "Images also include a lot of common sense information about the world. People learn this by themselves and, with NEIL, we hope that computers will do so as well."
To date, NEIL, which has been running since July 15 of this year, has analyzed over three million images.
More detailed findings will be published by the research team at the IEEE International Conference on Computer Vision in Sydney, Australia to be held on Dec. 4.
However, the researchers warn that the program is not 100 percent accurate. Errors might occur while recognizing and the goal cannot be achieved without the involvement of humans.
"People don't always know how or what to teach computers," Gupta said. "But humans are good at telling computers when they are wrong."
The project is being funded by Google and the Office of Naval Research.
