Structure-Aware Procedural Text Generation From an Image Sequence

It is an important activity for our society to create new value by combining materials. From daily cooking manufacturing industry, we often describe the way do it as a procedural text. As pointed some previous studies natural language understanding, one property of text its dependency context, which merging operations materials and can be represented graph or tree structure. This paper aims investigate impact explicitly introducing such structure on vision task generation from image sequence. To this end, propose (1) dataset, extends definition version (2) novel structure-aware model, learns context efficiently. Experimental results show that proposed method boost performance traditional versatile methods.