A Deep Learning based Architecture for Personal A.I. Fashion Stylist Services

Information

Title A Deep Learning based Architecture for Personal A.I. Fashion Stylist Services
Authors
Sungchan Park, Jeeseung Han, Jun Yeob Kim, Holim Lim, Sungjun Kim, Jeawoong Jung, Eunyoung Park, Sang-goo Lee, Yuri Lee, Jong-Youn Rha
Year 2019 / 11
Keywords fashion, artificial intelligence, deep learning, recommendation, software architecture
Acknowledgement KOCCA
Publication Type International Conference
Publication The 2nd Artificial Intelligence on Fashion and Textile International Conference (AIFT 2019)

Abstract

Despite its scope and glamourous contents, the fashion industry has been relatively low tech (in terms of IT) and slow to move online. The nuances of style and finely segmented products make keyword-oriented search inadequate, resulting in ineffective and often inaccurate recommendations. In response to such shortcomings, many attempts are being made to integrate big data analytics and social recommendations into fashion e-commerce. Further, there has been a steady increase in applications of deep learning oriented visual intelligence for a more refined search and user experience in the fashion domain. In this talk, we introduce a new personalized A.I. (artificial intelligence) stylist service that provides the user with daily outfit recommendations as well as shopping suggestions based on personal style and the inventory of clothing item she/he owns. We also present the data and system architectures for this service based on a novel deep learning model that is capable of vectorising the subtle style features of fashion item images. Utilizing these style vectors, the system supports visual search as well as automatic tagging of garment images. It can also compose outfits based on matching semantics learned from millions of outfit samples from the Internet - an outfit is an ensemble of garments and fashion accessories that make up one full wearing set.