Training new detection models aren’t supported yet.

However, you can train a recognition model to recognize persons as far as you have enough images to train.