Annotation

Annotation: This paper investigates adaptive level-of-detail control in 3D Gaussian Splatting (3DGS) under limited computational budgets. We propose a frame-wise Gaussian selection method that estimates primitive importance in screen space and formulates computational load control as a budget-constrained optimization problem. The method combines a screen-space cost estimate with a hybrid predictor that captures both spatial and temporal image characteristics. To ensure consistent evaluation, we develop a reproducible experimental framework for collecting unified performance and quality metrics, including frame rate, rendering time, and image quality measures. The proposed approach is evaluated against gradient-based selection strategies on scenes of varying complexity. The results show that using screen-space cost as the selection criterion yields a more robust trade-off between image quality and performance than gradient-based alternatives while remaining suitable for real-time applications. The proposed problem formulation, reproducible experimental setup, and unified set of metrics can be used in the development of standardized procedures for the comparative evaluation of adaptive rendering methods.