Leveraging Class Hierarchy in Fashion Classification

Abstract

The online commerce market has been growing rapidly, spurring interest in the deep fashion domain from the research community. Among various tasks in the fashion domain, the classification problem is the vital one, because metadata extraction through fashion classification has tremendous industrial value. A flurry of recent deep-learning based models have been proposed for the task and have showed great performances but they fail to capture the hierarchical nature of fashion annotations, such as 'pant' and 'skirt' both having 'bottom' as the superordinate. In this preliminary work, we propose a novel fashion classification model that works in a hierarchical manner. Experimental results on large fashion datasets show that our intuition, taking into account hierarchical dependencies between class labels, can help improve performance.