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:20251218T030657Z
LOCATION:Meeting Room S421\, Level 4
DTSTART;TZID=Asia/Hong_Kong:20251216T145000
DTEND;TZID=Asia/Hong_Kong:20251216T150000
UID:siggraphasia_SIGGRAPH Asia 2025_sess124_papers_1094@linklings.com
SUMMARY:TrackerSplat: Exploiting Point Tracking for Fast and Robust Dynami
 c 3D Gaussians Reconstruction
DESCRIPTION:Daheng Yin and Isaac Ding (Simon Fraser University); Yili Jin 
 (McGill University, Simon Fraser University); Jianxin Shi (Nankai Universi
 ty, Simon Fraser University); and Jiangchuan Liu (Simon Fraser University)
 \n\nRecent advancements in 3D Gaussian Splatting (3DGS) have demonstrated 
 its potential for efficient and photorealistic 3D reconstructions, which i
 s crucial for diverse applications such as robotics and immersive media.\n
 However, current Gaussian-based methods for dynamic scene reconstruction s
 truggle with large inter-frame displacements, leading to artifacts and tem
 poral inconsistencies under fast object motions.\nTo address this, we intr
 oduce \textit{TrackerSplat}, a novel method that integrates advanced point
  tracking methods to enhance the robustness and scalability of 3DGS for dy
 namic scene reconstruction.\nTrackerSplat utilizes off-the-shelf point tra
 cking models to extract pixel trajectories and triangulate per-view pixel 
 trajectories onto 3D Gaussians to guide the relocation, rotation, and scal
 ing of Gaussians before training.\nThis strategy effectively handles large
  displacements between frames, dramatically reducing the fading and recolo
 ring artifacts prevalent in prior methods.\nBy accurately positioning Gaus
 sians prior to gradient-based optimization, TrackerSplat overcomes the qua
 lity degradation associated with large frame gaps when processing multiple
  adjacent frames in parallel across multiple devices, thereby boosting rec
 onstruction throughput while preserving rendering quality.\nExperiments on
  real-world datasets confirm the robustness of TrackerSplat in challenging
  scenarios with significant displacements, achieving superior throughput u
 nder parallel settings and maintaining visual quality compared to baseline
 s.\nThe code is available at \href{https://github.com/yindaheng98/TrackerS
 plat}{https://github.com/yindaheng98/TrackerSplat}.\n\nRegistration Catego
 ry: Full Access, Full Access Supporter\n\nSession Chair: Yan-Pei Cao (VAST
 )\n\n
END:VEVENT
END:VCALENDAR