On augmenting database schemas by latent visual attributes

Abstract Decision-making in our everyday lives is surrounded by visually important information. Fashion, housing, dating, food or travel are just a few examples. At the same time, most commonly used tools for information retrieval operate on relational and text-based search models which well understood end users, but unable to directly cover visual contained images videos. Researcher communities have been trying reveal semantics of multimedia last decades with ever-improving results, dominated success deep learning. However, this does not close gap model its own often rather solves very specialized task like assigning one pre-defined classes each object within closed application ecosystem. Retrieval based these novel techniques difficult integrate existing application-agnostic environments built around databases, therefore, they so widely industry. In paper, we address problem closing between database model. We propose formalize discovering candidates new attributes analysis available content. design implement system architecture supporting attribute extraction, suggestion acceptance processes. apply solution context e-commerce show how it can be seamlessly integrated SQL last, evaluate user study discuss obtained results.