What does a good photo of food look like? Can a machine tell? Can we fix up photos of food using deep learning?