Yeah, I mean, the tagging is awesome, but I'm thinking I'll need more image segmentation than object recognition. With a segmentation map, I can make a great image->image translator.