Graphics Reference
In-Depth Information
•
To mathematically describe a large set of computer vision principles and
algorithms that underlie the tools used on a daily basis by visual effects artists.
•
To collect and organize many exciting recent developments in computer
vision research related to visual effects. Most of these algorithms have only
appeared in academic conference and journal papers.
•
To connect and contrast traditional computer vision research with the real-
world terminology, practice, and constraints of modern visual effects.
•
To provide a compact and unified reference for a university-level course on
this material.
This topic is aimed at early-career graduate students and advanced, motivated
undergraduate students who have a background in electrical or computer engi-
neering, computer science, or applied mathematics. Engineers and developers of
visual effects software will also find the topic useful as a reference on algorithms, an
introduction to academic computer vision research, and a source of ideas for future
tools and features. This topic is meant to be a comprehensive resource for both the
front-end artists and back-end researchers who share a common passion for visual
effects.
This topic goes into the details of many algorithms that form the basis of commer-
cial visual effects software. For example, to create the fight scene we just described,
we need to estimate the 3D location and orientation of a camera as it moves through
a scene. This used to be a laborious process solved mostly through trial and error by
an expert visual effects artist. However, such problems can now be solved quickly,
almost automatically, using visual effects software tools like
boujou
, which build
upon structure frommotion algorithms developed over many years by the computer
vision community.
On the other hand, this topic also discusses many very recent algorithms that
aren't yet commonplace in visual effects production. An algorithmmay start out as a
university graduate student's idea that takesmonths to conceive and prototype. If the
algorithm is promising, its description and a few preliminary results are published
in the proceedings of an academic conference. If the results gain the attention of a
commercial software developer, the algorithm may eventually be incorporated into
a new plug-in or menu option in a software package used regularly by an artist in
a visual effects studio. The time it takes for the whole process — from initial basic
research to common use in industry — can be long.
Part of the problem is that it's difficult for real-world practitioners to identifywhich
academic research is useful. Thousands of new computer vision papers are published
each year, and academic jargon often doesn't correspond to the vocabulary used to
describe problems in the visual effects industry. This topic ties these worlds together,
“separating thewheat fromthe chaff” and clarifying the research keywords relevant to
important visual effects problems. Our guiding approach is to describe the theoretical
principles underlying a visual effects problem and the logical steps to its solution,
independent of any particular software package.
This topic discusses several more advanced, forward-looking algorithms that
aren't currently feasible for movie-scale visual effects production. However, com-
puters are constantly getting more powerful, enabling algorithms that were entirely
impractical a few years ago to run at interactive rates on modern workstations.
