From the examples given, it can recognize the one face you show it as the one face you trained it on.
Even the premise is backwards, the face of the superhero is the known quantity, you should train the model on that, then use it to try and pick the super hero out of the pictures of everyone else in the show/movie. If nothing else, it wold provide negative examples show there's some chance of ruling out false positives.
Even the premise is backwards, the face of the superhero is the known quantity, you should train the model on that, then use it to try and pick the super hero out of the pictures of everyone else in the show/movie. If nothing else, it wold provide negative examples show there's some chance of ruling out false positives.