SIParCS 2015- Chu

Dongliang Chu, The University of Memphis

GPU-based Raycasting of Volumetric Data

(Slides)  (Recorded Talk)

Volume visualization of structured grids, arising primarily from numerical simulation, is an important technique for exploring data sets in numerous scientific disciplines. Volume visualization algorithms are computationally expensive, but readily parallelizable. Moreover, many volumetric rendering techniques lend themselves to highly efficient implementation on Graphics Processing Units (GPUs). Ray casting, in particular, maps well onto today’s highly flexible and powerful GPUs. However, most GPU based ray casting implementations are limited to regularly structured grids. Less work has been done in the area of irregularly structured grids, such as the curvilinear grids widely used in atmospheric and related sciences.
Curvilinear grids present a couple of major challenges to conventional, fixed-rate-sampling ray casting algorithms. First, curvilinear grids may have non-planar faces that complicate the ray-cell intersection calculation. Secondly, curvilinear grids such as those used in atmospheric and ocean modeling typically have widely varying cell sizes, making it difficult to ensure that each cell is adequately sampled.  Within this project we prototyped a GPU based ray casting algorithm that is suitable for curvilinear grids. Our algorithm uses adaptive sampling to ensure that each cell is properly sampled. Our implementation uses the OpenGL Shading Language. We evaluated both the accuracy and efficiency of our algorithm against a conventional GPU ray caster.

Mentors: John Clyne and Alan Norton, CISL