Supplementary material to "Top-down Visual Saliency Guided by Captions" (CVPR 2017) - View it on GitHub