Photo identification is done with a limited search space of “tiles” that the image is decomposed into, for convolutional NNs.