Investigating depth of field in volume rendering and distributed volume rendering on high performance computing systems

Update Item Information
Title Investigating depth of field in volume rendering and distributed volume rendering on high performance computing systems
Publication Type dissertation
School or College College of Engineering
Department Computing
Author Grosset, Pascal
Date 2016
Description The aim of direct volume rendering is to facilitate exploration and understanding of three-dimensional scalar fields referred to as volume datasets. Improving understanding is done by improving depth perception, whereas facilitating exploration is done by speeding up volume rendering. In this dissertation, improving both depth perception and rendering speed is considered. The impact of depth of field (DoF) on depth perception in direct volume rendering is evaluated by conducting a user study in which the test subjects had to choose which of two features, located at different depths, appeared to be in front in a volume-rendered image. Whereas DoF was expected to improve perception in all cases, the user study revealed that if used on the back feature, DoF reduced depth perception, whereas it produced a marked improvement when used on the front feature. We then worked on improving the speed of volume rendering on distributed memory machines. Distributed volume rendering has three stages: loading, rendering, and compositing. In this dissertation, the focus is on image compositing, more specifically, trying to optimize communication in image compositing algorithms. For that, we have developed the Task Overlapped Direct Send Tree image compositing algorithm, which works on both CPU- and GPU-accelerated supercomputers, which focuses on communication avoidance and overlapping communication with computation; the Dynamically Scheduled Region-Based image compositing algorithm that uses spatial and temporal awareness to efficiently schedule communication among compositing nodes, and a rendering and compositing pipeline that allows both image compositing and rendering to be done on GPUs of GPU-accelerated supercomputers. We tested these on CPU- and GPU-accelerated supercomputers and explain how these improvements allow us to obtain better performance than image compositing algorithms that focus on load-balancing and algorithms that have no spatial and temporal awareness of the rendering and compositing stages.
Type Text
Publisher University of Utah
Subject Depth of Field; Distrubuted Volume Rendering; HPC; Image Compositing; Perception
Dissertation Name Doctor of Philosophy in Computing
Language eng
Rights Management ©Pascal Grosset
Format application/pdf
Format Medium application/pdf
Format Extent 1,862,124 bytes
Identifier etd3/id/4195
ARK ark:/87278/s67d63jh
Setname ir_etd
ID 197741
Reference URL https://collections.lib.utah.edu/ark:/87278/s67d63jh
Back to Search Results