Smart target point control for Gaussian Splatting methods

2026-05-15

AI Summary

This work addresses the uncontrolled growth of Gaussian primitives in 3D Gaussian splatting, which stems from heuristic densification and pruning strategies and adversely affects reconstruction quality, computational efficiency, and fair method comparison. To resolve this, the authors propose an intelligent target-point control mechanism that dynamically adjusts the hyperparameters governing densification and opacity-based pruning while preserving the standard densification schedule. This enables the number of Gaussians to follow a smooth, predefined quadratic trajectory during training, avoiding abrupt truncation. For the first time, the method achieves precise control over Gaussian count throughout optimization, reliably reaching the target scale within 15k iterations. Consequently, different views and methods undergo balanced optimization under identical capacity constraints, significantly enhancing reconstruction consistency and evaluation reliability.

Abstract

Standard Gaussian splatting methods rely on heuristic densification and pruning to adaptively allocate primitives during training, and the resulting Gaussian count strongly influences both reconstruction quality and runtime. This makes comparisons across methods fragile: improvements can stem from higher representational capacity rather than algorithmic design. A common and naive workaround for this is hard-stopping or budgeting densification/pruning once a target count is reached, which biases training because different methods hit the cap at different times, yielding non-uniform densify/prune exposure across views and uneven point distributions. We propose a target point control scheme that preserves the standard densification window and cadence, but adjusts only the existing densification and opacity-culling hyper-parameters to track a quadratic target count trajectory. This quota-governor reaches the desired count by 15k iterations without abrupt cutoffs, ensuring that all methods and views receive equal densification and pruning cycles, enabling fairer, capacity-matched evaluation.

Problem

• Gaussian Splatting
• densification
• pruning
• target point control
• fair evaluation

Innovation

• Gaussian Splatting
• target point control
• adaptive densification
• fair evaluation
• opacity culling