Visual Semantic Relatedness Dataset for Image Captioning