Researchers propose TSOG to shrink 4D Gaussian Splatting by 90%

A team of researchers has proposed **TSOG (Temporally and Spatially Ordered Gaussians)**, a lossy compression format designed to dramatically reduce the storage footprint of 4D Gaussian Splatting (4DGS) content. Developed by Shady Gmira, Evangelos Alexiou, Emmanouil Potetsianakis, and Emmanuel Thomas, TSOG extends the Spatially Ordered Gaussians (SOG) framework into the temporal domain by introducing a timeline attribute and temporal parameterization for geometry and appearance attributes. Each Gaussian in TSOG is assigned a unique index, with attribute values encoded as index-aligned image data, similar to SOG.

In evaluations using a PLYs sequence and FreeTimeGS as baselines, TSOG achieved file size reductions exceeding 90% compared to simplistic and state-of-the-art 4DGS representations, respectively. The quality degradation was minimal, with PSNR differences ranging between -0.42 and +0.85 dB. The format is model-agnostic, extensible, and compatible with both discrete and continuous 4DGS representations, making it a promising solution for efficient representation, storage, and delivery of dynamic scenes for next-generation 4D content.