Proceedings: GI 2010

Memory efficient ray tracing with hierarchical mesh quantization

Benjamin Segovia, Manfred Ernst

Proceedings of Graphics Interface 2010: Ottawa, Ontario, Canada, 31 May - 2 June 2010, 153-160

  • BibTex

    @inproceedings{Segovia:2010:,
    author = {Segovia, Benjamin and Ernst, Manfred},
    title = {Memory efficient ray tracing with hierarchical mesh quantization},
    booktitle = {Proceedings of Graphics Interface 2010},
    series = {GI 2010},
    year = {2010},
    issn = {0713-5424},
    isbn = {978-1-56881-712-5},
    location = {Ottawa, Ontario, Canada},
    pages = {153--160},
    numpages = {8},
    publisher = {Canadian Human-Computer Communications Society},
    address = {Toronto, Ontario, Canada},
    }

Abstract

We present a lossily compressed acceleration structure for ray tracing that encodes the bounding volume hierarchy (BVH) and the triangles of a scene together in a single unified data structure. Total memory consumption of our representation is smaller than previous comparable methods by a factor of 1.7 to 4.8, and it achieves performance similar to the fastest uncompressed data structures. We store quantized vertex positions as local offsets to the leaf bounding box planes and encode them in bit strings. Triangle connectivity is represented as a sequence of strips inside the leaf nodes. The BVH is stored in a compact quantized format. We describe techniques for efficient implementation using register SIMD instructions (SSE). Hierarchical mesh quantization (HMQ) with 16 bits of accuracy achieves an average compression rate of 5.7: 1 in comparison to a BVH and an indexed face set. The performance impact is only 11 percent for packet tracing and 17 percent for single ray path tracing on average.