Facebook's Facial Recognition Nearing Human-Level Performance

 Facial Recognition  

Facebook's new AI Group has been working on a deep learning project called DeepFace, to develop facial recognition software which maps 3D facial features allowing facial recognition from any angle.

Research at the newly organized Facebook AI Group has already yielded a system that recognizes faces almost as well as a human. Called DeepFace, the work is not yet a part of the Facebook system, but promises incredible possibilities for photographic recognition and other applications.

"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."

Related articles Digital Avatars Crossing The Uncanny Valley  Researchers Use Network of Cameras to Track People Indoors  AI System Breaks CATCHA Turing Test

The system has reached 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 (determined for the same dataset at a mean of 97.5%).

So far DeepFace remains purely a research project. Facebook has released a research paper on the project and the researchers will present the work at the IEEE Conference on Computer Vision and Pattern Recognition in June.

DeepFace uses two processes to recognize faces. First it corrects the angle of a face so that the person in the picture faces forward, using a default 3D model of an forward-looking face. This means the faces do not have to be in a fixed orientation to be recognized by the system.

Next the deep learning algorithms are applied using a simulated neural network works to create a numerical description of the reoriented face. If DeepFace comes up with similar enough descriptions from two different images, it decides they must show the same face.

The researchers have trained DeepFace on the largest facial dataset to-date, an identity labeled dataset of four million facial images belonging to more than 4,000 identities, where each identity has an average of over a thousand samples.

This dataset has allowed for the accurate model-based alignment with the large facial database generalize remarkably well to faces in uncontrolled situations.

The performance of the final software was tested against a standard data set that researchers use to benchmark face-processing software, which has also been used to measure how humans fare at matching faces.

The study authors conclude:

Our work demonstrates that coupling a 3D model-based alignment with large capacity feedforward models can effectively learn from many examples to overcome the drawbacks and limitations of previous methods. The ability to present a marked improvement in face recognition, which is a central field of computer vision that is both heavily researched and rapidly progressing, attests to the potential of such coupling to become significant in other vision domains as well.

Neeraj Kumar, a researcher at the University of Washington who has worked on face verification and recognition, told MIT Technology Review that Facebook’s results show how finding enough data to feed into a large neural network can allow for significant improvements in machine-learning software. “I’d bet that a lot of the gain here comes from what deep learning generally provides: being able to leverage huge amounts of outside data in a much higher-capacity learning model,” he says.

With augmented reality increasingly becoming part of computer systems and hardware, facial recognition has the potential to be increasingly ubiquitous as well.  Systems like DeepFace hint at just how accurate such systems may be, and also point to a future where anonymity and privacy may be difficult or even impossible to maintain.


SOURCE  Facebook

By 33rd Square

Embed