|
Fabrice NEYRET - Maverick team, LJK, at INRIA-Montbonnot |
The field of Computer Graphics has some Graals such as
photorealism with complex light effects and materials, construction
and rendering of very detailed scenes, real-time exploration of very
large scenes, amplification (beautification) of coarse fluid or light
simulations, seamless merging of controled and automatic data, etc.
The realistic realtime walk-trough detailed galaxies somehow gathers
all of these. Galaxies inter-twin a « fluid »
of heterogeneous stars and fractal opaque filaments of dust clouds
which hide, semi-hide or are illuminated by star clusters or singular
stars.
A key strategy to tackle such mass-based data while
preserving real-time and realism is the design of scalable lazy (i.e.
minimal) representations and algorithms, able to encode directly the
visual phenomenas emerging from the sub-pixel scale. An other one is
to generate details on the fly from coarse data and statistical
informations.
Our GigaVoxels platform offers a convenient
framework for such scalable real-time exploration.
This subject addresses the second strategy mentioned above, and
aims at amplifying our galactic simulation with volumetric details
reproducing observations (consisting of statistical descriptions and
reference images). In particular, we target dust clouds multiscale
shape and its filaments, and the continuous zoom from density to
fluctuations, discrete clusters, then discrete stars.
Procedural
noise techniques such as Perlin noise and hypertextures are a natural
source of inspiration, but do not ensure any conservation of
quantities nor offers a transition from continuum to discrete. The
expected « noise » model will.
Math : basic stats and signal (Fourier) useful for describing images & textures properties would be a plus.
C/C++, OpenGL.
Notions of parallelism (multi-threading) or GPU programming (GLSL, CUDA, etc) would be a plus, but these can be learn easily during the practice.