BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Asia/Hong_Kong
X-LIC-LOCATION:Asia/Hong_Kong
BEGIN:STANDARD
TZOFFSETFROM:+0800
TZOFFSETTO:+0800
TZNAME:HKT
DTSTART:19911015T033000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20251218T030656Z
LOCATION:Meeting Room S426+S427\, Level 4
DTSTART;TZID=Asia/Hong_Kong:20251215T132000
DTEND;TZID=Asia/Hong_Kong:20251215T133100
UID:siggraphasia_SIGGRAPH Asia 2025_sess106_papers_1990@linklings.com
SUMMARY:NeLiF: Neural Lighting Function Generation for Real-Time Indoor Re
 ndering
DESCRIPTION:Hongtao Sheng (State Key Laboratory of CAD&CG, Zhejiang Univer
 sity); Yuchi Huo (State Key Laboratory of CAD&CG, Zhejiang University; Zhe
 jiang Lab); Chuankun Zheng, Guangzhi Han, and Bin Zang (State Key Laborato
 ry of CAD&CG, Zhejiang University); Yifan Peng (The University of Hong Kon
 g); Shi Li (State Key Laboratory of CAD&CG, Zhejiang University); Hao Zhu,
  Rui Tang, and Yiming Wu (Manycore Tech Inc.); and Rui Wang and Hujun Bao 
 (State Key Laboratory of CAD&CG, Zhejiang University)\n\nRecent advances i
 n neural rendering have extensively explored modeling\nthe radiance fields
  with neural representations, while overlooking the under-\nlying mechanis
 ms for producing various lighting effects, and consequently\nleading to li
 mited adaptability to dynamic scenes. These lighting effects,\nincluding h
 ighlights, shadows, and indirect lighting, are usually computed\nusing phy
 sically-based rendering methods like path tracing, which can be\ncomputati
 onally prohibitive for complex indoor luminaires. Although sev-\neral rece
 nt studies have attempted to model global illumination effects with\nneura
 l representations, they commonly suffer from long training times or\npoor 
 generalizability to novel scenes. In light of these challenges, this work\
 npresents a novel neural lighting function generation model that can syn-\
 nthesize diverse lighting effects in real time for unseen dynamic scenes a
 nd\ncomplex indoor luminaires, with results comparable to state-of-the-art
  ren-\ndering pipelines. Our model specifically consists of two stages. In
  the first\nstage, multi-view observation images of the luminaire are capt
 ured and\nthen used to encode a compact, scene-independent 3D neural light
 ing field.\nIn the second stage, light information is sampled from the neu
 ral lighting\nfield and combined with the G-buffers and shadow clues to pr
 oduce the\nshading results. In parallel, we leverage a state-of-the-art ge
 nerative model\ntogether with our HDR Lift module to generate an HDR 3D Ga
 ussian representation of the luminaire.In our experiments, the model train
 ed on a dataset of 10,000 modern indoor scenes demonstrates strong general
 izability, high efficiency, and visually convincing results across a wide 
 range of test scenes, highlighting its potential as a practical and flexib
 le solution for high-fidelity, real-time neural indoor rendering.\n\nRegis
 tration Category: Full Access, Full Access Supporter\n\nSession Chair: Yuc
 hi Huo (Zhejiang University, Korea Advanced Institute of Science and Techn
 ology)\n\n
END:VEVENT
END:VCALENDAR