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:20260817T171531Z
LOCATION:Hall B5 (2)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241206T104500
DTEND;TZID=Asia/Tokyo:20241206T105600
UID:siggraphasia_SIGGRAPH Asia 2024_sess143_papers_1185@linklings.com
SUMMARY:Gaussian Surfel Splatting for Live Human Performance Capture
DESCRIPTION:Zheng Dong (State Key Laboratory of CAD&CG, Zhejiang Universit
 y); Ke Xu (City University of Hong Kong); Yaoan Gao, Hujun Bao, and Weiwei
  Xu (State Key Laboratory of CAD&CG, Zhejiang University); and Rynson W.H.
  Lau (City University of Hong Kong)\n\nHigh-quality real-time rendering us
 ing user-affordable capture rigs is an essential property of human perform
 ance capture systems for real-world applications. However, state-of-the-ar
 t performance capture methods may not yield satisfactory rendering results
  under a very sparse (e.g., four) capture setting. Specifically, neural ra
 diance field (NeRF)-based methods and 3D Gaussian Splatting (3DGS)-based m
 ethods tend to produce local geometry errors for unseen performers, while 
 occupancy field (PIFu)-based methods often produce unrealistic rendering r
 esults. In this paper, we propose a novel generalizable neural approach to
  reconstruct and render the performers from very sparse RGBD streams in hi
 gh quality. The core of our method is a novel point-based generalizable hu
 man (PGH) representation conditioned on the pixel-aligned RGBD features. T
 he PGH representation learns a surface implicit function for the regressio
 n of surface points and a Gaussian implicit function for parameterizing th
 e radiance fields of the regressed surface points with 2D Gaussian surfels
 , and uses surfel splatting for fast rendering. We learn this hybrid human
  representation via two novel networks. First, we propose a novel point-re
 gressing network (PRNet) with a depth-guided point cloud initialization (D
 PI) method to regress an accurate surface point cloud based on the denoise
 d depth information. Second, we propose a novel neural blending-based surf
 el splatting network (SPNet) to render high-quality geometries and appeara
 nces in novel views based on the regressed surface points and high-resolut
 ion RGBD features of adjacent views. Our method produces free-view human p
 erformance videos of 1K resolution at 12 fps on average. Experiments on tw
 o benchmarks show that our method outperforms state-of-the-art human perfo
 rmance capture methods.\n\nRegistration Category: Full Access, Full Access
  Supporter\n\nLanguage Format: English Language\n\nSession Chair: Iain Mat
 thews (Epic Games, Carnegie Mellon University)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_1185&sess=sess143
END:VEVENT
END:VCALENDAR