Memory-Efficient Sliding Window Progressive Meshes

Abstract

Progressive mesh is a data structure that encodes a continuous spectrum of mesh approximations. Sliding window progressive meshes (SWPM) minimize data transfers between CPU and GPU by storing mesh data in static on-GPU memory buffers [For01]. The main disadvantages of the original SWPM algorithm are poor vertex cache usage efficiency, and big resulting datasets. Connectivity-based algorithm [KT04] achieves a good vertex cache coherence but it does not address the problem of high memory utilization. In this paper, we describe estimates for the size of memory buffers, and describe methods to reduce the index datasets. We achieve 20% reduction due to the use hierarchical data structures (clustered patches); further reduction (50% or more) is possible if one can optimize connectivity of input meshes, or is willing to decrease the number of mesh approximations stored in the progressive mesh.