There are no references to a paper or code. It isn't even clear to me that the example images represent unseen and unused data. The model could have just memori

There are no references to a paper or code. It isn't even clear to me that the example images represent unseen and unused data. The model could have just memorized the answers. A more realistic test of the model would be to plug in images not found in the training set.