Presentation

<
3DGS2: Near Second-order Converging 3D Gaussian Splatting
SessionSplatting Bigger, Faster, and Adaptive
DescriptionThis paper introduces a nearly second-order convergent training algorithm for 3D Gaussian Splatting that exploits independent kernel attributes and sparse coupling across images. By constructing and solving small Newton systems for parameter groups, it achieves about an-order faster training while maintaining or exceeding SGD-based reconstruction quality.
Authors
Lei Lan
University of Utah
State Key Lab of CAD and CG, Zhejiang University
Tianjia Shao
Zhejiang University
Zixuan Lu
University of Utah
Yu Zhang
University of Utah
Chenfanfu Jiang
UCLA
Yin Yang
University of Utah
Event Type
Technical Paper
TimeWednesday, 13 August 202510:45am - 10:55am PDT
LocationWest Building, Rooms 301-305
ACM Digital Library Journal Papers' PDFs
Conference Papers' PDFs
Session Time & Location
Sunday, 10 August 20256:00pm - 8:45pm PDTWest Building, Ballroom AB
Wednesday, 13 August 202510:45am - 12:35pm PDTWest Building, Rooms 301-305
Recordings
Livestreamed
Recorded
Next PresentationNext PresentationFLoD: Integrating Flexible Level of Detail into 3D Gaussian Splatting for Customizable Rendering
Similar Presentations
JGS2: Near Second-order Converging Jacobi/Gauss-Seidel for GPU Elastodynamics
3D Gaussian Splatting for Realistic Physical AI Simulations
Elastic Locomotion with Mixed Second-order Differentiation