Compression algorithm of point-based model based on multi-space arithmetic coding

QSplat is a typical point-based rendering system. Yet, though it has very compact data structure, its data sets are still very huge, necessary for further compression. We divide a node of QSplat model into 3 or 4 sections. For each section, a separated arithmetic coding space is created. In each space, 1 context-based adaptive arithmetic coding is respectively used to compress the data sets. We achieve a better compression rate and less memory usage than typical compression tools, and our algorithm can be a nice complement to QSplat.