BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Australia/Melbourne
X-LIC-LOCATION:Australia/Melbourne
BEGIN:DAYLIGHT
TZOFFSETFROM:+1000
TZOFFSETTO:+1100
TZNAME:AEDT
DTSTART:19721003T020000
RRULE:FREQ=YEARLY;BYMONTH=4;BYDAY=1SU
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:19721003T020000
TZOFFSETFROM:+1100
TZOFFSETTO:+1000
TZNAME:AEST
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20240214T070250Z
LOCATION:Meeting Room C4.11\, Level 4 (Convention Centre)
DTSTART;TZID=Australia/Melbourne:20231215T135600
DTEND;TZID=Australia/Melbourne:20231215T141100
UID:siggraphasia_SIGGRAPH Asia 2023_sess137_papers_130@linklings.com
SUMMARY:GroomGen: A High-Quality Generative Hair Model Using Hierarchical 
 Latent Representations
DESCRIPTION:Technical Communications, Technical Papers\n\nYuxiao Zhou (ETH
  Zürich), Menglei Chai and Alessandro Pepe (Google Inc.), Markus Gross (ET
 H Zürich), and Thabo Beeler (Google Inc.)\n\nDespite recent successes in h
 air acquisition that fits a high-dimensional hair model to a specific inpu
 t subject, generative hair models, which establish general embedding space
 s for encoding, editing, and sampling diverse hairstyles, are way less exp
 lored. In this paper, we present GroomGen, the first generative model desi
 gned for hair geometry composed of highly-detailed dense strands. Our appr
 oach is motivated by two key ideas. First, we construct hair latent spaces
  covering both individual strands and hairstyles. The latent spaces are co
 mpact, expressive, and well-constrained for high-quality and diverse sampl
 ing. Second, we adopt a hierarchical hair representation that parameterize
 s a complete hair model to three levels: single strands, sparse guide hair
 s, and complete dense hairs. This representation is critical to the compac
 tness of latent spaces, the robustness of training, and the efficiency of 
 inference. Based on this hierarchical latent representation, our proposed 
 pipeline consists of a strand-VAE and a hairstyle-VAE that encode an indiv
 idual strand and a set of guide hairs to their respective latent spaces, a
  hybrid densification step that populates sparse guide hairs to a dense ha
 ir model, and a neural simulator that deforms hair driven by head pose. Gr
 oomGen not only enables novel hairstyle sampling and plausible hairstyle i
 nterpolation, but could also enable interactive editing of complex hairsty
 les, or serve as strong data-driven prior for hairstyle reconstruction fro
 m images. We demonstrate the superiority of our approach with qualitative 
 examples of diverse sampled hairstyles and quantitative evaluation of gene
 ration quality regarding every single component and the entire pipeline.\n
 \nRegistration Category: Full Access\n\nSession Chair: Weidan Xiong (Shenz
 hen University)
URL:https://asia.siggraph.org/2023/full-program?id=papers_130&sess=sess137
END:VEVENT
END:VCALENDAR