Tutorials: Lossy Compression for Scientific Data: Principles, Tools, and Implications for Visualization
Organizers: Franck Cappello, Peter Lindstrom, Sheng Di, Robert Underwood, Hanqi Guo
Large-scale simulations, observations, and experiments generate massive scientific datasets that pose significant challenges for storage, transfer, and interactive visualization. Lossy compression is an effective technique to reduce data size while preserving essential scientific information. This tutorial introduces the visualization community to the principles, state-of-the-art tools, and implications of lossy compression for scientific data. We first cover the motivation and use cases of lossy compression, the underlying techniques (decorrelation, quantization, and coding), and the leading compressors, including SZ, ZFP, MGARD, and SPERR. We then focus on topics particularly relevant to the visualization community: how lossy compression affects visualization quality and feature preservation, error assessment metrics, hands-on exercises with compression tools, and customization of compressors using the FZ framework. The tutorial is presented by the leading researchers in scientific data compression, who have collectively developed the most widely used compressors in this domain. This tutorial builds on highly rated tutorials given at SC17—25 and ISC17—22, and is adapted here for the IEEE VIS audience, with emphasis on visualization-specific considerations.