How do I train a CNN to learn bounding boxes from labeled images?

Question

I am trying to detect faces of specific people in images. Two people. I have 2k images of each of the two people that are labelled. These are normal snapshots, so there are