Subscriber Discussion

Deepface: Facial Recognition At 97.25% Facebook Project

While it still exists in papers, it does look like we might be at the end of the tunnel for facial recognition, or is it?

Deepface: "Asked whether two unfamiliar photos of faces show the same person, a human being will get it right 97.53 percent of the time. New software developed by researchers at Facebook can score 97.25 percent on the same challenge, regardless of variations in lighting or whether the person in the picture is directly facing the camera."

Well, first this is not facial recognition. They explicitly note:

"Facebook’s new software, known as DeepFace, performs what researchers call facial verification (it recognizes that two images show the same face), not facial recognition (putting a name to a face)."

After that, they say it could be used for facial recognition but that's a big jump in level of difficulty. A lot easier to say if 2 things are the same, then what of 7 billion people a face belongs to.

Nearly 5 years, ago, 3VR trumpeted 90%+ real time facial recognition performance. Whatever happened to that? Whatever happened to the other vendors touting such claims?

I am sure things will get better, if simply because we can throw a lot more processing power today than 5 years ago making algorithms feasible now that would have been unrealistic to attempt in the past. That said, it is still slow going.