Prompt-Based Fashion Outfits Retrieval System

Abstract

The exponential growth of e-commerce in recent years has transformed the fashion industry, propelling it into a new era of digital retail. With the convenience of online shopping, consumers now have access to an extensive array of fashion products from the comfort of their homes and as a result in need of more efficient and personalized shopping experiences. This demand paved the way for the advancement of recommendation and retrieval systems in fashion e-commerce. In this thesis, we build a system plan to streamline and enhance the retrieval of fashion outfits from vast and diverse collections. Our system consists of two components, a multimodal model to retrieve image items matching a textual description, and a network incorporating hashing modules to capture high-order interactive compatibility between fashion items, facilitating efficient and personalized fashion outfit recommendations. Through extensive experimentation and evaluation, we demonstrate the effectiveness of our system in providing accurate and personalized fashion outfit recommendations with desired descriptions by the consumers, like a particular color, style, occasion, season, and many more