Turbo Learning for CaptionBot and DrawingBot

Neural Information Processing Systems 

We study in this paper the problems of both image captioning and text-to-image generation, and present a novel turbo learning approach to jointly training an image-to-text generator (a.k.a.