ControlGS: Consistent Structural Compression Control for Deployment-Aware Gaussian Splatting

Fengdi Zhang1,2, Yibao Sun2, Hongkun Cao2, Ruqi Huang1
1Tsinghua University, 2Pengcheng Laboratory
arXiv Code Models

TL;DR: ControlGS introduces a cross-scene consistent structural compression control mechanism for 3DGS optimization. By adjusting the control hyperparameter 𝜆𝛼, ControlGS can flexibly generate models biased toward either structural compactness or high fidelity, regardless of the specific scene scale or complexity, suitable for cross-device automated deployment.

Interactive Demo

Adjust 𝜆𝛼 (Lambda) or use “Auto” to auto-select the optimal model based on device performance.

Note: Browser rendering of 3DGS scenes may differ. Refer to the official viewer for accurate results.

Method Overview

ControlGS Method Overview

Starting from a sparse point cloud reconstructed via SfM, ControlGS initialize an Gaussian set and alternate between Uniform Splitting and pruning based on Opacity-based Sparsification:

  • Uniform Splitting: Whenever the number of pruned Gaussians falls below a threshold, we split all surviving Gaussians into eight children in an octree fashion, inheriting parent attributes to achieve a coarse-to-fine frequency progression without scene-specific heuristics.
  • Opacity-based Sparsification: We add an L1 opacity regularization that applies a constant negative gradient to each Gaussian’s opacity, causing low-contribution Gaussians to “self-shrink” and be removed once their opacity drops below a threshold.

The controllable core of ControlGS is a single hyperparameter, 𝜆𝛼, which scales the L1 loss:

  • Increasing 𝜆𝛼 strengthens pruning, yielding more structural compact models;
  • Decreasing 𝜆𝛼 preserves more Gaussians for higher-fidelity rendering.

Results

Structural Compression Control Performance

ControlGS achieves continuous, scene-agnostic, and highly responsive control over the trade-off between model structural compactness and rendering quality. The figure below shows how the control hyperparameter 𝜆𝛼 (x-axis) consistently influences rendering quality (y-axis) and the number of Gaussians (dot size) across diverse scenes, ranging from small objects to large outdoor scenes. For more detailed results, please refer to the paper.


Structural compression control performance

Comparison with competing method

ControlGS achieves higher rendering quality with fewer Gaussians on unseen test views, consistently preserving intricate structures and high-frequency textures across diverse scenes. The number of Gaussians used is indicated at the bottom-right corner of each subfigure below. For more detailed results, please refer to the paper.


Qualitative comparison with competing method

BibTeX

    @article{zhang2025controlgs,
      title={ControlGS: Consistent Structural Compression Control for Deployment-Aware Gaussian Splatting},
      author={Zhang, Fengdi and Sun, Yibao and Cao, Hongkun and Huang, Ruqi},
      journal={arXiv preprint arXiv:2505.10473},
      year={2025}
    }