It’s easy for humans to identify faces in pictures on Facebook, but the method isn’t as simple for computers. Sure, Facebook has a suggested prompt that predicts who you’re trying to tag, but now the company is working on a technology that promises “near-human accuracy” so you won’t have to do it yourself in the future.
Facebook’s API Group is developing software called DeepFace, which maps 3D facial features and creates a colorless model to narrow in on specific characterizations. The accuracy of the method is 97.25%, which is just under the 97.5% accuracy that a human can identify, according to the group.
“We present a system (DeepFace) that has closed the majority of the remaining gap in the most popular benchmark in unconstrained face recognition, and is now at the brink of human level accuracy,” researchers said in a report released by Facebook API Group. “It is trained on a large dataset of faces acquired from a population vastly different than the one used to construct the evaluation benchmarks, and it is able to outperform existing systems with only very minimal adaption.”
To develop the technology, Facebook looked at 4.4 million tagged faces from 4,030 of its users to help the system learn how to better identify features specific to each person. The report also reveals that Facebook looks at modern face recognition in four phases: detect, align, represent and classify.
“We revisit both the alignment step and the representation step by employing explicit 3D face modeling in order to apply a piecewise affine transformation, and derive a face representation from a nine-layer deep neural network,” the company notes on its DeepFace page.
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 4,000 identities, where each identity has an average of over a thousand samples.
Although we might not see the updated approach on Facebook just yet, the site is expected to present it at the IEEE Conference on Computer Vision and Pattern Recognition in June, according to MIT Technology Review.
Source: Mashable