The DeepFace project by Facebook
Facebook has revolutionised the world of social networking. The founders of the social networking site began it in their rooms while at university in Harvard, and what began as a simple thing has grown exponentially. They were recently in the news for acquiring the messaging service whatsapp, with speculation rife as to the direction they are taking. Recently, they have invented software that they claim can match faces better than even humans do. According to research, a human being is 97.53 per cent likely to match photos of two unfamiliar faces. This is regardless of the lighting situation or whether the people in the pictures face the camera.
Early we write about a dangerous from social networks side but Facebook and its competitors have been actively developing and betting on the software called deep learning. It is an artificial intelligence software that uses a complex method of matching faces. The software makes use of a network of neurons that help it slowly learn to recognise patterns in large pools of data. The new software scores 97.25 per cent in recognising faces, which is a significant improvement over past artificial intelligence software. According to Facebook's artificial intelligence researchers, it closely approaches human performance. The error margin has been reduced by more than a quarter relative to earlier software.
The software uses a 3 dimensional model that rotates faces, doing it virtually such that even if they are facing away they appear to be facing the camera. It does what researchers call facial verification as opposed to facial recognition. Facial verification involves recognizing whether two images show the same face while in facial recognition, software puts names to faces. The two techniques can be tied together, therefore improving the ability of Facebook to recognise whether two faces are the same and put names to them. Connecting these techniques together could be helpful in suggestions of whom to tag in photos but this technology may be also dangerous.
The DeepFace project, which is what Facebook calls it, is in the preliminary stages of research, at least for the moment. Facebook recently published a paper on the project aiming at the technology conference by the IEEE in June. The aim of making the publication will be to get feedback from other members of the technological research community. The software pioneers are Yaniv Taigman together with Ming Yang and another colleague Marc'aurelio Ranzato. Also involved in the project development was a professor Lior Wolf from a university in Israel. Taigman has been speaking in the public circles a lot about the project. We in Jammer-store.com hope that this project will be directed to develop in a positive direction.
The DeepFace project works in two stages. In the first stage, the software virtually changes the angle of the picture. The correction is such that the picture faces forward in the picture. The software uses a 3 dimensional model of a typical average front facing face. In the next stage, the neural network comes in. It works out a numerical description of the face. In this stage, it compares the numerical data collected from the two images. If there are enough similarities in the data, then the software decides that the two images are of the same face. If you don`t trust this modern technology we recommend to choose portable or stationary jammers which we offer in our store.