University of Washington is testing facial recognition software to see which one is the most accurate when being used on larger scale.

Researchers observe the algorithms on facial recognition software - which previously was proved to perform its job accurately. However, the algorithms were being used for only 13,000 image dataset.

300 groups of researchers conduct MegaFace challenge to observe the accuracy of facial recognition software

The researchers tried to use the facial recognition program to a larger scale of image collections. They used a million person scale by collecting images of people around the world from Flickr. There were more than 690,500 pictures being used to test the system.

The program should be able to determine if two images were the exact same individual. The result claimed that the accuracy drops when the program is used on a larger scale, Silicon Republic reported.

Researchers suggest developing more accurate algorithms

Testing on small scale does not determine the algorithm accuracy. It would have more reliable result when used on bigger scale. The test included facial recognition program developed by Google and N-TechLab.

FaceNet - a facial recognition software developed by Google - is said to have the highest accuracy rate with 75 percent on the million person dataset. Previously, FaceNet was also being tested on smaller scale and the result was almost perfect.

N-TechLab, a Russia developed facial recognition software - also shows a drop in accuracy rate after being used on million person scale test. Google FaceNet, N-TechLab and Beijing's Faceall are the highest-three software to have better accuracy than the others. All of the results are displayed at MegaFace Challenge official website

The findings suggest improvement on the AI software to ensure its effectiveness when identifying a face. University of Washington is still conducting the challenge till the end of the month.

Related
Tags University of washington