MirrorGAN: Learning Text-To-Image Generation by Redescription | IEEE Conference Publication | IEEE Xplore