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:20260817T171530Z
LOCATION:Hall B5 (2)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241206T110800
DTEND;TZID=Asia/Tokyo:20241206T111900
UID:siggraphasia_SIGGRAPH Asia 2024_sess143_papers_1054@linklings.com
SUMMARY:GGHead: Fast and Generalizable 3D Gaussian Heads
DESCRIPTION:Tobias Kirschstein, Simon Giebenhain, and Jiapeng Tang (Techni
 cal University of Munich); Markos Georgopoulos (Independent); and Matthias
  Nießner (Technical University of Munich)\n\nLearning 3D head priors from 
 large 2D image collections is an important step towards high-quality 3D-aw
 are human modeling. \nA core requirement is an efficient architecture that
  scales well to large-scale datasets and large image resolutions. \nUnfort
 unately, existing 3D GANs struggle to scale to generate samples at high re
 solutions due to their relatively slow train and render speeds, and typica
 lly have to rely on 2D superresolution networks at the expense of global 3
 D consistency. \nTo address these challenges, we propose Generative Gaussi
 an Heads (GGHead), which adopts the recent 3D Gaussian Splatting represent
 ation within a 3D GAN framework. \nTo generate a 3D representation, we emp
 loy a powerful 2D CNN generator to predict Gaussian attributes in the UV s
 pace of a template head mesh. \nThis way, GGHead exploits the regularity o
 f the template's UV layout, substantially facilitating the challenging tas
 k of predicting an unstructured set of 3D Gaussians. \nWe further improve 
 the geometric fidelity of the generated 3D representations with a novel to
 tal variation loss on rendered UV coordinates. \nIntuitively, this regular
 ization encourages that neighboring rendered pixels should stem from neigh
 boring Gaussians in the template’s UV space. \nTaken together, our pipelin
 e can efficiently generate 3D heads trained only from single-view 2D image
  observations. \nOur proposed framework matches the quality of existing 3D
  head GANs on FFHQ while being both substantially faster and fully 3D cons
 istent. \nAs a result, we demonstrate real-time generation and rendering o
 f high-quality 3D-consistent heads at 1024x1024 resolution for the first t
 ime.\n\nRegistration Category: Full Access, Full Access Supporter\n\nLangu
 age Format: English Language\n\nSession Chair: Iain Matthews (Epic Games, 
 Carnegie Mellon University)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_1054&sess=sess143
END:VEVENT
END:VCALENDAR