This is a partial implementation of Jason Weber and Joseph Penn's paper, "Creation and Rendering of Realistic Trees", from SIGGRAPH 1995. This code should not be treated as an example on how to implement that paper properly (even though there are some confusing sections in the paper that I've tried to interpret in a useful fashion), but more as a basic example on how one could set up nested object instance hierarchies to maximize memory efficiency in PRMan 17.0 (as described in my User's Group Meeting talk at SIGGRAPH 2012).

Sections 4.1 through 4.5 of the Penn/Weber paper are mostly implemented, 4.6 through 4.9 are not. I also did not implement ternary stem splits, handling of -nCurveV (helical stems), or trunk lobes.

To use this plugin, compile it as a DSO. On Linux, this can be done using a command resembling:

g++ -I$RMANTREE/include -fPIC -o arbre.so -shared arbre.cpp

The resulting shared object can be loaded with RiProcedural2 using a recent version of prman:

Procedural2 "DynamicLoad" "SimpleBound" "float[6] bound" [-20 20 -20 20 0 20] "string dsoname" ["arbre"] "int Seed" [66] "int MaxLevel" [5] "int StartLevel" [0] "int InstanceLevel" [4] "int Leaves" [1] "int InstanceLeaves" [1]

The useful parameters are:

"string Type": tree type. Currently supported: oak, tupelo, and aspen. The parameters are straight out of the paper, slightly modified in a few cases

"int Seed": random seed for the tree. If you are growing a forest of trees, every tree should have its own seed

"int MaxLevel": maximum level of the tree - this overrides the Levels parameter in the paper

"int InstanceLevel": level at which to use instanced branches. For example, if InstanceLevel == MaxLevel, the terminal level of the tree will use instanced branches. For two levels of terminal instanced branches, set InstanceLevel == MaxLevel - 1, and so on. To disable instanced branches, set this parameter to 0.

"int Leaves": toggle switch for whether leaves will be grown or not

"int InstanceLeaves": whether leaves will be instanced. If 1, the procedural requires a "leaf" ObjectInstance to have been previously defined. If 0, the procedural requires a "leaf" inline archive to be previously defined.

"int StartLevel": by default, the procedural starts at Level 0, this parameter overrides that default. This is mainly useful for visualization purposes

There are several known bugs with this plugin that ought to be addressed before even considering using this for production work.

• If a stem has high curvature, its children may be detached at the base. This is "fixed" by switching to polygons. The proper fix for subdivision meshes is to interpolate using the cubic b-spline rather than just perform linear interpolation as is done here

• The actual formulae for the leaf/branch frequency/stem length/stem radius are dependent not only parent properties but also grandparent properties. So, the simple-minded instancing being used here where we just look for the branch of the right length isn't close to correct as far the Penn/Weber model is concerned. I've attempted to compensate for this by associating an arbitrary parent offset with each stem length, and this seems to be "good enough" at a first glance. A more rigorous derivation, coupled perhaps by searching for appropriate instances in multiple dimensions (e.g. radius and length), may provide more accurate results.

• The use of erand48 throughout is completely ad hoc and wacky.

Naturally, there are other useful features that should be implemented as well (like switching amongst multiple leaf types).

Julian Fong [email protected]