Micro-Polygon vs. Texture Map Debate
CS348c - Computer Graphics Architectures
June 4, 1996
Overview of Micro-Polygon and Texture Mapping Techniques
Micro-Polygons and Texture Mapping are two techniques for providing
detail in 3d Computer Graphics. In the micro-polygon approach, the scene
is composed of many small polygons, each of which will project to one screen
space pixel or less. This composition could be either at the modeling level,
or a scene can be decomposed to micro-polygons as part of the rendering
phase. In texture mapping multi-pixel polygons are used, but complex images
are mapped onto their surfaces to give the appearance of extra complexity.
The two techniques together define a spectrum of rendering styles, from
a QuickTime VR type system where the entire scene is essentially one polygon
with an image mapped onto it, all the way to the point where there are
many sub-pixel polygons each of which essentially has a one pixel texture
map.
For the purposes of this debate, however, we will consider micro-polygon
to refer to 2-pixel or smaller sized polygons, and texture mapping to refer
to Reality Engine style rendering, with triangles in the 25+ pixel size
range.
Suggested Reading
So far there are two papers for the debate:
Debate Structure
The benefits and/or disadvantages of the two techniques will be debated
around the following issues:
- Economics - for a given level of detail, which gives a better
price/performance
- Architectural Complexity - how complicated is it to implement
the different techniques
- Performance - which technique performs more quickly
- Maximum Detail Capacity - Which technique generates the most
realistic looking images (for real vs. synthesized scenes)?
Overview Slides
Here
are the overview slides for the debate (by Brad Johanson).
Micro-Polygons:
General Concessions:
Micropolygon's advantages lie in rendering scenes of large complexity
and high resolution. I will concede that in situations of numerous large
polygons and low complexity the Micropolgon way does not fair well.
However, the trends are the following:
- Larger scene complexity Computing power combined with consumer expectations
is pushing up the bar in terms of what is "acceptable" image
quality - which is a direct factor of scene complexity.
- Decreased polygon size Triangles size used to be assumed that it was
100 pixels. Currently, triangles are assumed to be around 25 pixels in
size. The Micropolygon approach uses polygons between 1/2 pixel and 2 pixels
large.
Remember that when comparing the two systems in Economics you must consider
detail/performance vs. cost. Saying a Texture Mapped system is faster if
it is of lower detail is meaningless.
Thus, it is my contention that as scene complexity continues to increase
and polygon size decreases, the advantages of using Micropolygons will
severely outweigh the "traditional" R.E. style polygon based
texture map.
Economic and Architectural Issues, Mike
Schroepfer
- One look into the architecture of the R.E. or IR shows that a large
multitude of the work involved is used in the rasterization process (80
processors per board) - as compared to 8- 12 geometry processors for the
entire system.
- Much of this work involves texturing - But since you are texturing
first, in object coordinates if you use pre-filtered mit-mapped textures
no filtering or interpolation is needed.
- Everything is essentially broken down into the same elements - the
Micropolygon. This could simply hardware.
- Takes advantage of locality both in textures and in geometry
This is an extreme advantage because low cost systems often skimp on
memory (DRAM based PC graphics boards). Even on high-end systems it is
much easier to get high performance out of serial memory accesses rather
than random one
What is currently Lacking in the Micropolygon Approach
Not nearly as much development time has been put into hardware specialized
to run these algorithms.
1. A hardware dicer
2. Methods to deal with the variety of higher level primitives.
The I.R. took longer with more effort than the R.E. It appears that
we may have to think in a new paradigm in order to continue the rapid advance
of graphic processing power. The simplicity of the back end of the Micropolygon
approach seems a possible way to do this.
Mike's Slides
Performance and Detail Issues, David
Redkey
Micropolygons out-perform standard hardware texture-mapping and generate
better images too!
- Detailed geometry is the only way to capture appropriate occlusion
and perspective distortion in realistic scenes. As Michael Deering concludes
in "Where do all the Triangles go", increasing numbers of small
polygons are needed to match the demand for realism.
- Micropolygons allow a clean framework for displacement maps and curved
surfaces, which will become prevalent in the future.
- Micropolygons allow for reduced super-sampling. As shading models become
more complex, the ratio of the costs of vertex transformation to sample
shading goes down, making the micropolygon approach more desirable.
- By designing hardware that focuses only on gouraud shading during rasterization,
we can certainly design faster rasterizers.
- Triangle bandwidth need not be a constraining factor if modern software
visibility techniques are employed.
David's
Micropolygon Slide David's
Performance Slide
Texture Maps:
Economic and Architectural Issues, Brien
Oberstein
Economics
- texture maps give a high level of detail at a lower cost
- more suitable for consumer graphics applications
- may be able to reuse texturing hardware for other effects
- combination of texture and bump maps good enough
Architectural Complexity
- textures are fairly straightforward
- textures add the appearance of a high image complexity without much
computation
- texture mapping can take advantage of simd
- micropolygon hardware implementation ???
Performance and Detail Issues, Li-Yi
Wei
Texture mapping and micropolygons are 2 way to capture realism. Basically
texture mapping is more hacking because it tries to capture the local and
global shading of objects via some cheaper way and at the same time produce
convincing pictures. The benefits for texture mapping compared to micropolygons
are :
- it's cheaper in terms of triangle bandwidth
- it's suitable for random access textures
- in applications that most objest are far away and are flat, texture
mapping is good enough, like flight simulator and Doom. Micropolygons can
give us better detail but at the expense of a large amount of tiny triangles
and also unsuitable for RATs.
PostScript of Debate Overhead for Texture Performance and Detail: tex_perf.ps
This page created by:
Brad Johanson
bjohanso@graphics.stanford.edu