The aim of the class projects is for each student/group to make some research
contribution to the field of rendering (widely defined). Of course,
things can go wrong, but you should at least have tried something with
research potential, even if it doesn't work out. You won't be penalized
if what seems like a reasonable initial hypothesis turns out to be false.
At the bottom of this page are a large number of suggested topics,
very briefly described. You can select one of those topics,
which I guarantee to know something about, or you can suggest your own,
which I may then approve. You will have a hard time convincing me to
allow something which doesn't involve a significant rendering angle.
I anticipate that you will work in pairs, but I will allow you to work
alone if you really want to.
Part-way through the project, the Monday before Thanksgiving, you will be
required to give a 20 minute in class presentation describing your progress
so far. That may include the results of a literature search, some
preliminary ideas, and/or some preliminary algorithm implementations.
The primary aim is to force you to start early, and to identify potential
At the end of the semester, you will be required to give an in class
presentation of your results. You will also be required to do a demo of
some kind (assuming there is something to demo). Everybody should see
everybody else's demo. Finally, you will be required to submit a short
(4-6 page) paper describing what you did, why, and how well it worked.
The lofty goal is to produce papers to submit to SIGGRAPH in early
- Fri Nov 5: Deadline for choosing a project
- Mon Nov 20: Progress reports presented in class
- Wed Dec 13: Project presentations in class
- Mon Dec 18: Papers and demo deadline
The project will be worth 70% of your final grade for the course. The grade
will depend on four factors:
- The progress report
- The final project presentation
- The project as demonstrated
- The project as documented
Some Suggested Project Topics
- Metropolis Sampling for Extended Form Factors The Metropolis
light transport algorithm has the disadvantage that it finely samples
diffuse illumination. Its strength is in finding and sampling difficult
paths. Apply Metropolis in a situation where its strengths are
used to best advantage, while avoiding its weaknesses. One such
application is in computing extended form factors (and also caustic
- Benchmarking Terrain LOD Algorithms There are several
algorithms available for rendering terrain in the form of height
fields at various levels of detail. There are, however, no established
benchmark terrains for those algorithms to be compared on. Implement
the algorithms, design the benchmarks, and run the tests.
- Automatic Landscape Generation There are ways to generate
terrain with particular random properties. But I am not aware of many
ways to automatically populate the terrain (with trees, houses, roads,
grass, etc), without significant user effort. Furthermore, for large
terrains it may be possible to generate the landscape on demand,
so that it need not be stored in full detail.
- Greater Coherence in NPR Animations As is frequently stated,
coherence is a problem in NPR animations. Something better can surely
be done, possibly by drawing on better probabilistic algorithms
for placing strokes.
- Cartoon Physics Rendering Cartoonists stretch and squish a
ball as it bounces. Generate such animation automatically. A sketch
at SIGGRAPH this year suggested one possible approach.
- NPR Contrast Manipulation Real paints and canvasses have real
limitations on the range of brightnesses and colors they can represent.
Artists work around this by manipulating the light and shade in a
composition in order to generate the perception of greater contrast
or brightness than there really is present in the painting. Vermeer
is an excellent example - he did it all the time. Furthermore,
technical illustration manuals suggest enhancing the contrast across
edges to highlight the boundary and accentuate depth. Try to develop
a way to capture this effect.
- Tone Reproduction Scenes computed with physically based
rendering algorithm have physically accurate luminance values
that must be represented with an image. Monitors and papers don't
have anywhere near the required dynamic range to reproduce the
required luminance values, so some algorithm must be used to squash
the range of brightnesses down to something that can be reproduced,
while attempting to generate the same visual impact as the
accurate brightness values. Such algorithms are tone reproduction
algorithms, and there is plenty of space to develop better ones.
This problem is strongly related to the one above.
Prof. Stephen Chenney
Back to CS838-3 Home Page