Research Analyzer
← Back ICRA 2026

GARO: Geometry-Aware Redundancy Optimization for Real-Time and High-Fidelity Dynamic Gaussian Splatting

Huiwen Xue, Kaixing Zhao, Zuheng MING, Tingcheng Li

PDF

AI summary

Key figure (auto-extracted from paper)
GARO achieves over 2× faster real-time rendering in dynamic scenes by intelligently pruning redundant Gaussians based on optimization activity and geometric curvature.
Dynamic Gaussian Splatting Redundancy Pruning Geometry-Aware Optimization Real-Time Rendering Novel View Synthesis 3D Reconstruction

Problem

Dynamic 3D Gaussian Splatting methods accumulate redundant Gaussians during training, leading to excessive memory usage and slow rendering speeds that hinder real-time applications.

Approach

The method introduces a unified redundancy measurement framework that identifies and removes redundant Gaussians by jointly evaluating their gradient-based optimization activity and local geometric curvature during adaptive density control.

Key results

  • Over 2× rendering speedup on synthetic and real-world datasets
  • Significant reduction in Gaussian count and GPU memory consumption
  • Maintains comparable or superior PSNR and SSIM reconstruction quality
  • Effectively prunes flat-region Gaussians while preserving high-curvature details

Why it matters

Enables efficient, real-time dynamic scene reconstruction for VR, robotics, and autonomous driving applications with limited computational resources.

Abstract

Novel view synthesis is a key task for dynamic scene reconstruction, where high rendering speed is essential for applications such as virtual reality. Existing deformable Gaussian Splatting methods achieve high-fidelity dynamic scene modeling, but still face limitations in memory usage and rendering efficiency due to the large number of redundant Gaussians. To address these challenges, we propose Geometry- Aware Redundancy Optimization (GARO), a unified redun- dancy measurement framework in the adaptive density control stage of the traditional dynamic scene reconstruction pipeline. This framework first selects low-gradient candidates using an optimization activity assessment strategy, and then evaluates geometric complexity through low curvature analysis to further filter and prune redundant points, resulting in a compact and expressive Gaussian representation. Extensive experiments on synthetic and real-world datasets demonstrate that GARO achieves robust trade-offs between quality and speed, with PSNR remaining stable and rendering speed improved by 2×, validating the efficiency and effectiveness of GARO.

Index terms

Deep Learning for Visual Perception Visual Learning Mapping

Related papers