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:20250110T023312Z
LOCATION:Hall B5 (1)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241204T111900
DTEND;TZID=Asia/Tokyo:20241204T113100
UID:siggraphasia_SIGGRAPH Asia 2024_sess112_papers_241@linklings.com
SUMMARY:PVP-Recon: Progressive View Planning via Warping Consistency for S
 parse-View Surface Reconstruction
DESCRIPTION:Technical Papers\n\nSheng Ye, Yuze He, Matthieu Lin, Jenny She
 ng, and Ruoyu Fan (Tsinghua University); Yiheng Han (Beijing University of
  Technology); Yubin Hu (Tsinghua University); Ran Yi (Shanghai Jiao Tong U
 niversity); Yu-Hui Wen (Beijing Jiaotong University); Yong-Jin Liu (Tsingh
 ua University); and Wenping Wang (Texas A&M University)\n\nNeural implicit
  representations have revolutionized dense multi-view surface reconstructi
 on, yet their performance significantly diminishes with sparse input views
 . A few pioneering works have sought to tackle the challenge of sparse-vie
 w reconstruction by leveraging additional geometric priors or multi-scene 
 generalizability. However, they are still hindered by the imperfect choice
  of input views, using images under empirically determined viewpoints to p
 rovide considerable overlap. We propose PVP-Recon, a novel and effective s
 parse-view surface reconstruction method that progressively plans the next
  best views to form an optimal set of sparse viewpoints for image capturin
 g. PVP-Recon starts initial surface reconstruction with as few as 3 views 
 and progressively adds new views which are determined based on a novel war
 ping score that reflects the information gain of each newly added view. Th
 is progressive view planning progress is interleaved with a neural SDF-bas
 ed reconstruction module that utilizes multi-resolution hash features, enh
 anced by a progressive training scheme and a directional Hessian loss. Qua
 ntitative and qualitative experiments on three benchmark datasets show tha
 t our framework achieves high-quality reconstruction with a constrained in
 put budget and outperforms existing baselines.\n\nRegistration Category: F
 ull Access, Full Access Supporter\n\nLanguage Format: English Language\n\n
 Session Chair: Michael Wimmer (TU Wien)
URL:https://asia.siggraph.org/2024/program/?id=papers_241&sess=sess112
END:VEVENT
END:VCALENDAR