PGMS: Pyramidal Gaussian Mixture Splatting for 3DGS Compression

PGMS introduces a plug-and-play 3DGS compression framework that reformulates dense Gaussian primitives into a rate-controllable pyramidal mixture representation. With hierarchical mixture clustering, rate-adaptive budget allocation, and a compositing-aware parameter update, it achieves strong compression and consistently outperforms prior 3DGS compression methods across standard benchmarks.

Recommended citation: Xiaoge Zhang, Zijie Wu, Mingtao Feng, Saeed Anwar, & Ajmal Mian. (2026). PGMS: Pyramidal Gaussian Mixture Splatting for 3DGS Compression.