AI Baidu was trained to recognize actions in video

AI Baidu was trained to recognize actions in video


The artificial intelligence program created by Chinese Internet giant Baidu, took part in the competition in which it was necessary to watch the videos and learn what is happening there. According to the results of the tests AI from Baidu ranked first, ahead of the program member from the second position by half a percent accuracy.

Competition ActivityNet was to demonstrate the capabilities of artificial intelligence for face recognition, objects and actions captured on video. Baidu has developed a facial recognition program, created as an alternative to tickets. She is able to find the person in the photo uploaded from social networks, and the control point decides whether the person pass through the turnstile. This program and it was decided to “send” on the competition.

During the competition the program was viewed more than 300 thousand short clips of different subjects, with 87 percent of them she described is true. For the recognition of objects AI from Baidu uses spatial-temporal modeling, which is used for training the neural network to the allocation of live colors and sounds. Once the desired clip is processed, the received data are processed and classified.

A similar system of recognition of persons and objects already tested in France and Japan, but to introduce large quantities of the development are not going first, they should finalize.