The distribution of light inside the scene (and inside plants) is computed using
a hierarchical radiosity algorithm based on instanciation. Indeed, to be able to cope with the
large geometric complexity of plant models, it has been essential to develop a technique which
memory cost is sub-linear in terms of the number of polygons in the scene.
Due to the high self similarity that exists in plant models, it is possible to represent a whole plant
as a collection of instances of a single part of it. This representation is only an approximation
because (to the difference of Fractals) self-similarity in plants is not exact; however is exists at
multiple scales (branches, leaves, whole plants, etc) and it is thus theoretically possible to
expand a plant using only a few basic elements. On the image below,
we show an example of such a representation:
The two images below show the difference between our multi-level instanciation method and a
classical hierarchical radiosity method with clustering. It appears clearly that our methos uses only a few
memory as compared to the other one and runs much faster.
Classical HR with clustering. 1 hour 57 mn, 123 MB of memory
Hierarchical instanciation. 26 mn, 13MB of memory.
The image below shows lighting simulations obtained using our hierarchical instanciation method.
Measurements were conducted on a Sgi Origin2000 workstation. In such cases our Instantiation algorithm prooves
very usefull because the entire model can not fit in the main memory of the machine.
Marron tree. 1 hour, 80 MB of memory
10 poplar trees of various ages. 2 hours, 83 MB of memory.
