Top-Down Visual Saliency Guided by Captions | IEEE Conference Publication | IEEE Xplore