Companies are making rapid process in facial recognition technology, developing hardware and software capable of distinguishing faces, emotions and movements in realtime. But all of that work is proprietary — how does the Open Source community keep up?
Researchers at Carnegie Mellon University have put together an open source facial recognition program based on Google’s FaceNet research. Called OpenFace, the developers say that it can recognize faces in real time with just 10 reference photos of the person.
We don't shill.
Check out TNW's Hard Fork.
Check out the video below to see the program in action:
The datasets that the developers used to train the program to recognize faces are relatively paltry compared to the training of a proprietary source, but they say results are promising:
Even though the public datasets we trained on have orders of magnitude less data than private industry datasets, the accuracy is remarkably high and outperforms all other open-source face recognition implementations we are aware of on the standard LFW benchmark.
While it still has a ways to go to be a thoroughly accurate and speedy facial recognition program, OpenFace is on the right track.
➤ OpenFace [GitHub]