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:20251218T104000
DTEND;TZID=Asia/Hong_Kong:20251218T105000
UID:siggraphasia_SIGGRAPH Asia 2025_sess152_papers_1801@linklings.com
SUMMARY:Gradient-Weighted Feature Back-Projection: A Fast Alternative to F
 eature Distillation in 3D Gaussian Splatting
DESCRIPTION:Joji Joseph, Bharadwaj Amrutur, and Shalabh Bhatnagar (Indian 
 Institute of Science)\n\nWe propose a training-free method for feature fie
 ld rendering in 3D Gaussian Splatting, enabling fast and scalable embeddin
 g of high-dimensional features into 3D scenes. Unlike training-based featu
 re distillation methods, which are computationally expensive and often yie
 ld feature embeddings that poorly reflect the rendered semantics, our appr
 oach back-projects 2D features onto pre-trained 3D Gaussians using influen
 ce weights derived from the rendering equation. This projection produces a
  queryable 3D feature field, validated on tasks including 2D and 3D segmen
 tation, affordance transfer, and identity encoding, spanning queries using
  language, pixel, and synthetic embeddings. These capabilities, in turn, e
 nable downstream applications in augmented and virtual reality, interactiv
 e scene editing, and robotics. Across different tasks, our method achieves
  performance comparable to or better than training-based approaches, while
  significantly reducing computational cost. The project page is at https:/
 /jojijoseph.github.io/3dgs-backprojection\n\nRegistration Category: Full A
 ccess, Full Access Supporter\n\nSession Chair: Lin Gao (University of Chin
 ese Academy of Sciences)\n\n
END:VEVENT
END:VCALENDAR