A possible optimization worth trying with this project would be to generate just half the face and then mirror it before running the facial detection fitness fu

A possible optimization worth trying with this project would be to generate just half the face and then mirror it before running the facial detection fitness function. Seeing as how your subject matter is generally symmetrical the algorithm as it is now is spending a lot of time just finding symmetrical images.