On Improving the Compression Efficiency of C3DGS

3D Gaussian Splatting (3DGS) enables real-time novel view synthesis but produces scene representations of several hundred megabytes, limiting practical deployment. This work proposes four targeted enhancements to the C3DGS compression pipeline: a softmax-based centroid weighting for sensitivity-aware clustering, two alternative reconstruction-error sensitivity measures, adaptive arithmetic coding, and an entropy-driven criterion for automatic codebook size selection. Each modification is evaluated independently. The combined model achieves approximately 22\% size reduction relative to C3DGS with negligible quality loss across standard benchmarks.