Symmetrically to Issue #1 , implement a simple way to annotate instances as appropriate or not. In this case, we want to annotate images as appropriate for training face identification models. Implement some simple rules, such as "exactly one face recognized", "not too small", "frontal".
