Master Thesis MSTR-2017-29

BibliographyGeringer, Sergej: Spatial CPU-GPU data structures for interactive rendering of large particle data.
University of Stuttgart, Faculty of Computer Science, Electrical Engineering, and Information Technology, Master Thesis No. 29 (2017).
131 pages, english.
Abstract

In this work, I investigate the interactive visualization of arbitrarily large particle data sets which ft into system memory, but not into GPU memory. With conventional rendering techniques, interactivity of visualizations is drastically reduced when rendering tens- or hundreds of millions of objects. At the same time, graphics hardware memory capabilities limit the size of data sets which can be placed in GPU memory for rendering. To circumvent these obstacles, a progressive rendering approach is employed, which gradually streams and renders all particle data to the GPU without reducing or altering the particle data itself. The particle data is rendered according to a visibility sorting derived from occlusion relations between different parts of the data set, leading to a rendering order of scene contents guided by importance for the rendered image. I analyze and compare possible implementation choices for rendering particles as opaque spheres in OpenGL, which forms the basis of the particle rendering application developed within this work. The application utilizes a multi-threaded architecture, where data preprocessing on a CPU-thread and a rendering algorithm on a GPU-thread ensure that the user can interact with the application at any time. In particular it is guaranteed that the user can explore the particle data interactively, by ensuring minimal latency from user input to seeing the effects of that input. This is achieved by favoring user inputs over completeness of the rendered image at all stages during rendering. At the same time the user is provided with an immediate feedback about interactions by re-projecting all currently visible particles to the next rendered image. The re-projection is realized with an on-GPU particle-cache of visible particles that is built during particle data streaming and rendering, and drawn upon user interaction using the most recent camera confguration according to user inputs. The combination of the developed techniques allows interactive exploration of particle data sets with up to 1.5 billion particles on a commodity computer.

Full text and
other links
Volltext
Department(s)University of Stuttgart, Institute of Visualisation and Interactive Systems, Visualisation and Interactive Systems
Superviser(s)Ertl, Prof. Thomas; Reina, Dr. Guido; Gralka, Patrick
Entry dateMay 28, 2019
   Publ. Computer Science