名词解析: fine-grained visual processing: VQA: stacked attention networks:论文找来看看 natural basis for attention:注意力的自然基础:Global Contrast based Salient Region Detection salient regions:最能表达图片主旨的区域 coco caption的四种评价: BLEU-4:分数越高越接近人类的判断 METEO