Fast implicit KD-trees

Many scientific data sets are 3D or 4D scalar fields, for which typically isosurface- and volume visualization methods are used to extract information. These data sets are either massively complex (e.g., seismic data sets), or steadily increasing in size due to the permanently improving resolutions of different 3D scanners (e.g., CT- and MRTscanners) or calculation results (e.g., CFD-simulations). Only algorithms that scale well to data set complexity are suited to visualize those increasing data sets. Isosurface ray tracing and maximum intensity projection (MIP) accelerated through implicit KD-trees have a logarithmic dependency between visualization time and scene size, making them ideal algorithms for the visualization of massively complex scalar fields. Furthermore is ray tracing efficiently parallelized on the more and more commonly used shared memory machines (e.g., desktop machines with several multicore processors) and may be used to realize advanced shading eff ects like shadows and reflections. We introduce new optimized implicit KD-trees which allow on today's desktop computers interactive isosurfacing and MIP of data sets that are bigger than one half of the machine's main memory.