AI-Stylist: An AI-based Framework for Clothing Aesthetic Understanding

Abstract

Current studies in computer science on clothing aesthetic evaluation are difficult to be practically applied as fashion is a subjective field affected by various factors, like cultural background, or physical information (e.g. skin color). On the other hand, computer is good at processing logical tasks with clear definition and evaluation index. In this work, we propose a framework which divides the aesthetic evaluation task into three levels, namely apparel harmony, personal compatibility, and personality expression, to reduce its complexity such that computer has great potential to evaluate the clothing aesthetic practically. Each level is clearly defined, and the technic solutions is proposed correspondingly. Initial work with examples is presented to demonstrate the potential of the proposed framework.