Apple’s tech is only geared towards authenticating a single face on specific controlled hardware. With that as its goal, Apple has branched away from traditional FR into an Apple ecosystem situation that makes them stand alone, apart from the greater FR industry.
FR’s strength is observing large numbers of faces, with those faces being a wide range of distances from the camera, comparing against large galleries of people while providing real-time responses to a variety of external systems. Apple’s implementation has none of these requirements.
Additionally, traditional FR is often deployed in an installation; meaning the camera is in a fixed location. When a camera is in a fixed location, it can exploit that fact with additional in-place learning to better mitigate the specific illumination of the camera’s view, as well as illumination transitions from the time of day and seasonal changes. Apple’s FR can’t-do any of this. It has to do some tracking of the subtle changes in a person’s face from variations of consumption and aging, but nothing like what traditional FR has to do when it compensates for the same person from a different camera, different view, different illumination, different makeup/shave – and handle those variations across millions and millions of people simultaneously.
A quality FR system works with any quality video, any quality camera, because they have to: the real world has a wide variety of video qualities and a wide variety of security cameras and non-security intended cameras that already exist, and current FR systems have to work thru them if those companies want to sell to corporations with existing security camera networks.
The iPhoneX is reported to “only take 500ms” to authenticate a face. That is embarrassingly long for any modern FR system. But then again, they don’t have to be fast as they only have a gallery of one face. The FR software I develop takes 25ms to authenticate a face, and we do 25 million face comparisons per second. That is the type of performance a deployed FR system has to exhibit because it has to compare large person galleries against crowds such as building entrances in real time.
Plus, the budgets for in the wild securities are often slim, so the computers are old and the cameras are old, but the results have to be cutting edge regardless.
Simply put, the greater FR industry has constraints and requirements Apple is simply not participating. So they stand apart, as a curiosity and consumer confuser.