Scalable
Geometry Roundtable
Scalable geometry' seemed
to be the buzzwords of the convention. This made attendance to anything
dealing with scalable geometry very good. The three scalable geometry
roundtable sessions were on both the programming and visual arts tracks,
but were attended almost entirely by programmers. The artists that attended
did provide an important point of view, but the discussions were mostly
technical in nature.
Since there wasn't enough time to cover topics in detail, the discussions
were generally kept at a conceptual level. The two hottest topics were
Multi-Resolution-Meshes (of which the most well known type is a progressive
mesh) and splines of various forms. Since many of the same subjects
were visited several times throughout the various sessions the data
below reflects a merging of all three hours. In no particular order,
a list of topics covered and some of the ideas discussed follows.
Multi-Resolution-Meshes
Creation
Error Quadrics and/or Energy Preservation (progressive mesh)
The general consensus was that Error Quadrics were the best method to
use. They are faster, easier to implement and generate better meshes
for most implementations and models.
Data Structures
Tree structures, node elements, and methods for creation were discussed
several times. Binary trees were the only stated implementation, which
follows with edge collapsing being the only method used to generate
the trees. Some discussion of data elements needed for view-dependent
mesh refinement also took place.
Texture Coordinate and Other Material Preservation Problems
The material boundary preservation method used for Error Quadrics was
shown. This method involves placing a boundary plane perpendicular to
the surfaces at a material boundary. Methods for texture coordinate
preservation ranged from locking down vertices to weighting child vertex
coordinates to re-projecting textures for every vertex in the tree.
Artistic Control via Weighting and Clamping
The need for artistic control of the mesh reduction process was discussed
several times. The methods currently in use involve weighting and/or
clamping of vertices.
Simplification/Refinement
Non-View Dependent
Non-view dependent simplification methods in use today are generating
a mesh with a specific polygon count.
View Dependent
View dependent simplification methods attempt to refine a mesh only
in areas determined to be important from the given view point. Some
of the view parameters discussed are listed below.
Screen Size - This is maintained by comparing a bounding
volumes screen size to a given tolerance.
Silhouette Preservation- This is accomplished by comparing
a vertex normal cone to the view direction and refining only if a portion
of the cone shows that a child vertex may be a silhouette element.
Back Facing - This is similar to Silhouette Preservation,
but refining stops if the view cone is entirely back facing.
View Frustum Culling- The mesh is refined only if the bounding
volume is in the view frustum to avoid processing portions of a mesh
that are not visible.
Animation
How do you animate a continuous skinned MRM?
Very little work had been done in this area, but the approaches that
have been used and were discussed are very similar to traditional animation
methods.
Spline/Patch Meshes
Tessellation
Subdivision
This seems to have been the preferred method of patch tessellation in
shipping products to date. The problems of recursion associated with
this method were discussed.
Strip Creation
There was little experience in this area. The primary discussion revolved
around crack prevention.
Crack Prevention
Crack prevention by boarder preservation and interior resolution was
discussed several times. The prevention of cracks through subdivision
methods was also discussed.
See Terrain below
Terrain/Height Fields
Adaptive Tessellation of Height Fields
(Lindstrom paper from SIGGRAPH 96)
The Lindstrom paper was a very popular subject. It seems as though many
people have implemented a version of this technique. It was unclear
as to whether anyone had actually used this in production though.
Patch Based Terrain
Subdivision
Methods for terrain patch subdivision were discussed. In particular,
crack prevention during the subdivision of patches was discussed several
times.
H-Splines
There was some interest in the use of H-Splines for representing terrain.
Other than fractal based terrain modeling there was no experience with
this however.
Tools
Getting patch data
Extracting patch data from several modeling packages was discussed,
but no detailed information came out about any implementation.
Viewing or previewing MRMs
The need for an artist to preview a Multi-Resolution Mesh, especially
when artistic control parameters are available, was discussed.
