Thursday, 11 September 2008

AHM08: Visualising the Future

Chris Johnson: Scientific Computing & Imaging Institute, Utah .. pronounced 'ski'

Not retrospective visualisation of the results, but integrated visualisation in the problem solving process

GPU - massively parallel architecture
.. scaling many times faster than multi-core cpu
.. now have high precision floating pt gpu from nvidia

using GPUs to process petabytes of neuro slice data.

volume rendering ...
traditional 'maximum intensity projection' (MIP) to 'full volume rendering'

new was too computationally extensive, but with GPU becomes tractable +
multi-dimensional transfer function - mapping derivatives & integrals across multi slices to rgb
... s/w called seg3d .. bioimage .. hardest part was making it usefull !

time dependant visualisation

isosurface extraction
.. marching cubes Lorensen & Cline 1987
.. but ow pisa, rtrt, noise, octree up to 10^4 faster algorithm, but not available .. ie opensource,

pisa .. livnat & tricoche '04 .. if the triangle is too small to see, dont calc it.

ray-tracing
as # objects goes up ray tracing becomes more efficient than raster (traditional gpu) algorithm
DOE asci c-safe .. simulate explosion from first principles & vis it.
.. manta - real time ray tracer
.. how to simulate the right colours of flames correctly, rather than map temp to colour ramp
.. perception of shadows .. currently base on phong and gurow in 1970s but todays hardware is faster
.. so if solve maxwell for realism .. need to artificially introduce an appropriate light source into say cat scan .. not always obvious how to do it.

3d vis of error and uncertainty ..
.. working on it . no one way to do it
.. what about mapping rgb to fuzzyness or sensitivity or confidence
.. uncertainty animation

in 2003 as much info was generated as was published in all preceeding human history, and have repeated that every year since.

cf www.vistrails.org with taverna & myexperiment .. visualisation of differences due to technique variation.

http://www.sci.utah.edu/vaw2007/ .. book from Visualisation and Analytics Workshop

No comments: