Effective Translation Models for Heterogeneous Schemas of Visual Descriptions

As results of active studies and analyses of image/video data, there have been a number of visual descriptions that have their own schemas. To increase the effectiveness of visual descriptions with heterogeneous schemas, it is required to design careful translation models that serves as a bridge between visual descriptions. In this paper, we suggest two types of translation models, and describe their properties. As a case study we implement a translator between two visual descriptions to show how the translation is performed.