Face recogntion is considered to be a solved problem by many. If you restrict youselves to standard datasets, the accuracy is in high 90’s. But the caveat in this is the phrase ‘standard datasets’. Outside these datasets, the performance drops perceptably. A perfect example for this would be the accuracies in Megaface, which is in 70’s. But the question is that do we need such a high accuracy in many cases. Of course it is nice to have a 100% accurate algorithm. But many real life usecases, we can capture a small set of images or even a short video. This will increase the diversity of faces avaliable for recognition, and consequently allow us to reliably use the current algorithms in real world.