More about HKUST
Large-scale Gaussian Splatting
PhD Qualifying Examination
Title: "Large-scale Gaussian Splatting"
by
Mr. Yunfan ZENG
Abstract:
3D Gaussian Splatting (3DGS) has emerged as an explicit scene representation
for real-time novel view synthesis, combining ideas from point-based
rendering and neural radiance fields to achieve high visual quality with
efficient training and rendering. While 3DGS performs well on object-centric
and moderate-scale scenes, extending it to large, unbounded environments
introduces substantial challenges in memory consumption, optimization
scalability, and rendering efficiency.
In this survey, we provide a comprehensive review of large-scale Gaussian
splatting techniques. We first revisit the foundations of Gaussian splatting
and its relation to neural rendering. We then systematically examine recent
advances across representation design, including level-of-detail modeling,
compression, hybrid primitives, and scene synthesis. Furthermore, we discuss
scalable training strategies, GPU-oriented rendering optimizations, and
streaming pipelines that enable deployment in large-scale settings. Finally,
we outline open challenges and future research directions toward scalable and
practical 3D scene representations.
Date: Friday, 24 April 2026
Time: 3:30pm - 5:30pm
Venue: Room 2132C
Lift 22
Committee Members: Prof. Pedro Sander (Supervisor)
Prof. Long Quan (Chairperson)
Dr. Dan Xu