Adaptive maturation schedule to optimize bandwidth with informative Gaussians

Develop an adaptive maturation and update schedule for SwinGS’s sliding-window 3D Gaussian Splatting pipeline that optimizes bandwidth usage by prioritizing the replacement and updating of the most informative 3D Gaussians (e.g., those representing human motion) rather than updating a preassigned fixed subset, thereby better allocating per-frame transmission under bandwidth constraints.

Background

SwinGS currently matures Gaussians on a fixed schedule to enable efficient GPU batch operations. Although relocation and a “birth” field permit limited inter-slice migration, the update policy does not explicitly prioritize which Gaussians to transmit and update under a bandwidth budget.

The authors describe an ideal scenario where the system replaces the least useful Gaussians with the most informative new ones, such as prioritizing motion-related Gaussians over static background, but note that they have not yet achieved this bandwidth optimization.