LARGE SCALE VISUAL RECOGNITION OF CLOTHING, PEOPLE AND STYLES

Kiapour, Mohammadhadi

Download PDF

Request Version for Screen Reader

Last Modified

March 19, 2019

Creator

Kiapour, Mohammadhadi
- Affiliation: College of Arts and Sciences, Department of Computer Science

Abstract

Clothing recognition is a societally and commercially important yet extremely challenging problem due to large variations in clothing appearance, layering, style, body shape and pose. In this dissertation, we propose new computational vision approaches that learn to represent and recognize clothing items in images. First, we present an effective method for parsing clothing in fashion photographs, where we label the regions of an image with their clothing categories. We then extend our approach to tackle the clothing parsing problem using a data-driven methodology: for a query image, we find similar styles from a large database of tagged fashion images and use these examples to recognize clothing items in the query. Along with our novel large fashion dataset, we also present intriguing initial results on using clothing estimates to improve human pose identification. Second, we examine questions related to fashion styles and identifying the clothing elements associated with each style. We first design an online competitive style rating game called Hipster Wars to crowd source reliable human judgments of clothing styles. We use this game to collect a new dataset of clothing outfits with associated style ratings for different clothing styles. Next, we build visual style descriptors and train models that are able to classify clothing styles and identify the clothing elements are most discriminative in every style. Finally, we define a new task, Exact Street to Shop, where our goal is to match a real-world example of a garment item to the same exact garment in an online shop. This is an extremely challenging task due to visual differences between street photos that are taken of people wearing clothing in everyday uncontrolled settings, and online shop photos, which are captured by professionals in highly controlled settings. We introduce a novel large dataset for this application, collected from the web, and present a deep learning based similarity network that can compare clothing items across visual domains.

Date of publication