BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Asia/Tokyo
X-LIC-LOCATION:Asia/Tokyo
BEGIN:STANDARD
TZOFFSETFROM:+0900
TZOFFSETTO:+0900
TZNAME:JST
DTSTART:18871231T000000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260817T171532Z
LOCATION:Hall B5 (1)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241206T095800
DTEND;TZID=Asia/Tokyo:20241206T100900
UID:siggraphasia_SIGGRAPH Asia 2024_sess139_papers_273@linklings.com
SUMMARY:Hairmony: Fairness-aware hairstyle classification
DESCRIPTION:Givi Meishvili, James Clemoes, Charlie Hewitt, Zafiirah Hoseni
 e, Xian Xiao, Martin de La Gorce, Tibor Takacs, Tadas Baltrusaitis, Antoni
 o Criminisi, and Chyna McRae (Microsoft); Nina Jablonski (Pennsylvania Sta
 te University); and Marta Wilczkowiak (Microsoft)\n\nWe present a method f
 or prediction of a person's hairstyle from a single image. Despite growing
  use cases in user digitization and enrollment for virtual experiences, av
 ailable methods are limited, particularly in the range of hairstyles they 
 can capture. Human hair is extremely diverse and lacks any universally acc
 epted description or categorization, making this a challenging task. Most 
 current methods rely on parametric models of hair at a strand level.\nThes
 e approaches, while very promising, are not yet able to represent short, f
 rizzy, coily hair and gathered hairstyles. We instead choose a classificat
 ion approach which can represent the diversity of hairstyles required for 
 a truly robust and inclusive system. Previous classification approaches ha
 ve been restricted by poorly labeled data that lacks diversity, imposing c
 onstraints on the usefulness of any resulting enrollment system. We use on
 ly synthetic data to train our models. This allows for explicit control of
  diversity of hairstyle attributes, hair colors, facial appearance, poses,
  environments and other parameters. It also produces noise-free ground-tru
 th labels. We introduce a novel hairstyle taxonomy developed in collaborat
 ion with a diverse group of domain experts which we use to balance our tra
 ining data, supervise our model, and directly measure fairness. We annotat
 e our synthetic training data and a real evaluation dataset using this tax
 onomy and release both to enable comparison of future hairstyle prediction
  approaches. We employ an architecture based on a pre-trained feature extr
 action network in order to improve generalization of our method to real da
 ta and predict taxonomy attributes as an auxiliary task to improve accurac
 y. Results show our method to be significantly more robust for challenging
  hairstyles than recent parametric approaches. Evaluation with taxonomy-ba
 sed metrics also demonstrates the fairness of our method across diverse ha
 irstyles.\n\nRegistration Category: Full Access, Full Access Supporter\n\n
 Language Format: English Language\n\nSession Chair: Kui Wu (LIGHTSPEED)\n\
 n
URL:https://asia.siggraph.org/2024/program/?id=papers_273&sess=sess139
END:VEVENT
END:VCALENDAR