← All publications
MICRO, 20253DGS

GCC: A 3DGS Inference Architecture with Gaussian-Wise and Cross-Stage Conditional Processing

Minnan Pei, Gang Li, Junwen Si, Zeyu Zhu, Zitao Mo, Peisong Wang, Zhuoran Song, Xiaoyao Liang, Jian Cheng

Overview

GCC accelerates novel view synthesis from a reconstructed 3D scene by skipping unused Gaussians and reusing each Gaussian across image tiles. GCC uses cross-stage conditional processing to skip preprocessing for primitives that never affect the image. Its Gaussian-wise rendering dataflow finishes the contributions of one Gaussian before loading the next. The paper compares the speed and energy efficiency of this 3D Gaussian Splatting (3DGS) accelerator with GSCore.

三维重建得到高斯场景后,GCC 在渲染新视角时跳过用不到的高斯,并在图像分块间复用高斯数据,减少端侧计算与访存。GCC 面向端侧低功耗的新视角合成。跨阶段条件处理让三维高斯溅射(3DGS)加速器跳过无用的预处理,逐高斯渲染则减少重复读取。论文将这一硬件设计与 GSCore 比较,评估渲染性能与能效。

  • 3D Gaussian Splatting
  • Computer architecture
  • Conditional processing

MICRO, 2025

Research by Minnan Pei (裴旻楠) and collaborators · Email