Could it not just be really good are recreating the image it was trained on or is it generally doing this with novel images in this case?
Could it not just be really good are recreating the image it was trained on or is it generally doing this with novel images in this case?