Abstract
dc:descriptionStyling plays a crucial role in the fashion industry, and for years, there has been a search for a way to virtually visualize outfits on a person accurately and instantly. Virtual try-on (VTON) systems offer this capability, but several technical challenges have limited their widespread commercial adoption. These challenges fall into three main categories: (1) quality - generating an appealing and accurate try-on image; (2) intractability - handling outfit combinations, instant feedback, styling variations, model selection, etc.; (3) practicality - addressing product catalog coverage, logistics, and costs. In this thesis, we thoroughly examine the key components of virtual try-on systems and develop an effective solution to address these challenges. Our approach introduces a multi-garment VTON method that accurately captures garment attributes and provides instant responses. We also devise a technique that adjusts garment draping (e.g., tucked vs. untucked) while maintaining the garment's identity, allowing for the generation of rendering policies that dictate outfit styling and layering. Moreover, we suggest various enhancements to improve detail preservation and overall quality. Our proposed VTON system not only achieves remarkable results on benchmark datasets but has also been successfully implemented by top fashion retailers in the industry.
Degree
thesis:*- Name thesis:degree_name
- Ph.D.
- Level thesis:degree_level
- Dissertation
- Discipline thesis:degree_discipline
- Computer Science
- Grantor
- University of Illinois at Urbana-Champaign
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Li, Kedan
- Contributors dc:contributor
-
- Forsyth, David Alexander
- Lazebnik, Svetlana
- Schwing, Alexander Gerhard
- Farhadi, Ali
Subjects
dc:subject × 5Rights
dc:rights- Statement dc:rights
-
- Copyright 2023 Kedan Li
- Language dc:language
- en, eng
Identifiers
dc:identifier.*- Handle dc:identifier
- https://hdl.handle.net/2142/122215