Distinguish static-geometry opacity peaks from motion compensation

Determine whether a local opacity peak in Opacity-Guided Splitting and Growth Regulation (OSGR) arises from persistent static geometry or from compensation for object motion in dynamic Snapshot Compressive Imaging (SCI) measurements.

Background

The proposed OSGR densification strategy uses local opacity contrast as an additional cue for splitting Gaussian primitives. This cue is designed for the paper’s static-scene formulation, in which a single Gaussian representation explains all masked views multiplexed into one SCI measurement.

For dynamic scenes, independently moving objects violate the static forward model, so opacity increases may represent either persistent surface support or compensation for motion-induced residuals. The paper reports that dynamic scenes remain a limitation and notes that repeated opacity-guided splitting could allocate representation capacity to ambiguous regions, preserving ghost contours or fragmented geometry. The unresolved issue is therefore to identify the physical source of an observed local opacity peak before using it to guide refinement.