Smart target point control for Gaussian Splatting methods
This work addresses the problem of unfair comparisons in Gaussian splatting research by providing a method to control primitive count during training, ensuring equal densification cycles across methods.
Gaussian splatting methods rely on heuristic densification, making comparisons unfair due to varying Gaussian counts. The authors propose a target point control scheme that adjusts hyper-parameters to track a quadratic count trajectory, reaching the target by 15k iterations without abrupt cutoffs, enabling fairer capacity-matched evaluation.
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.