Facebook and Google can easily have millions of labeled face images, and there are now publicly available face databases of 0.5M face images, so training data: not an issue. All systems perform some preprocessing to align faces to a canonical frontal view, so any face image that is not totally a side view is usually well recognized. Strong pose (both tilt and pan), illumination changes and occlusions are still tricky though.
Curious about the issue of this lawsuit.
Slashdot Mirror
Check rankings on LFW: http://vis-www.cs.umass.edu/lf..., most of the best performing systems are using deep learning.
Facebook and Google can easily have millions of labeled face images, and there are now publicly available face databases of 0.5M face images, so training data: not an issue. All systems perform some preprocessing to align faces to a canonical frontal view, so any face image that is not totally a side view is usually well recognized. Strong pose (both tilt and pan), illumination changes and occlusions are still tricky though. Curious about the issue of this lawsuit.