Facebook’s DeepFace project is being considered as a new milestone in computer vision and facial recognition. Last week, Facebook announced that it had developed a new program called DeepFace, capable of identifying the face of a person with 97.25% accuracy which is at par with that of an average human when it comes to facial recognition.
This program has been designed by three in-house Facebook developers, Yaniv Taigman, Ming Yang and Marc’ Aurelio Ranzato and Lior Wolf, a professor at Tel Aviv university. As an example, the developers show that DeepFace can successfully recognise that this is Academy Award winner Sylvester Stallone.
Earlier, facial recognition via computer could be easily foiled if a subject is tilting their head in a slightly different direction. DeepFace program would be much more intensive and uses the software to correct the angle of a face in an image and then compares that to a 3D model of an average face. It then stimulates the neural network to find a numerical description of the face.
The existing Facebook’s facial recognition software suggests friends to tag when you upload a photo using information such as distance between eyes, nose and eyes in profile pictures and already tagged photos. These results are not as accurate as the DeepFace feature as it uses techniques which specialises in understanding irregular type of data.
As per Facebook, “This deep network involves more than 120 million parameters using several locally connected layers without weight sharing, rather than the standard convolutional layers. Thus, we trained it on the largest facial dataset to-date, an identity labeled dataset of four million facial images belonging to more than 4000 identies, where each identity has an average of over a thousand samples. The learned representations coupling the accurate model-based alignment with the large facial database generalise remarkably well to faces in unconstrained environments, even with a simple classifier. Our method reaches an accuracy of 97.25% on the Labeled faces in the Wild (LFW) dataset, reducing the error of the current state of the art by more than 25%, closely approaching human level performance.”
The business model of Facebook is based on understanding all the information we post on the social network, which leverages data to personalize ads so you’ll be more likely to click on them. As of now, this project is put forward as an academic pursuit, in a research paper released last week and the research team would present its findings at the Computer Vision and Pattern Recognition conference in Columbus, Ohio in June.