[Beowulf] Digital Image Processing via HPC/Cluster/Beowulf - Basics

Lux, Jim (337C) james.p.lux at jpl.nasa.gov
Tue Nov 6 08:12:06 PST 2012

I think the workflow for this kind of thing is definitely evolving over
time..  And there is a HUGE difference between high quality "feature
animation" (think Avatar) and games, music videos, old school Saturday
morning cartoons (the latter often derisively referred to as color xerox

One big factor is the number of polygons and the resolution of the images
being rendered.. High end feature animation uses a lot more polygons to
render someone's head or arms, for instance, and might use a more
sophisticated gridding of those polygons.  You also see things like hair
and drapery effects in a feature, which are less common in the game world
(hard to model, very computationally expensive to calculate the polygon or
facet positions in a dynamic sense, and back to the lots o' polygons
problem). There are also things like blur effects and such that get
tweaked and rendered.  I would imagine that the typical feature CG
animation has a whole raft of people who do nothing other than arrange
light sources (functions performed by gaffers and the Director of
Photography on a live action set).

If you look at the "extra features" stuff on DVDs for big animated
features from the 90s, you'll see the artist working at a workstation,
generally not a generic PC.  Today, yes, with fast GPUs and the software
that supports them, they probably are running on a PC of some sort.  But
they're probably running some sort of custom software that is tied to a
specific hardware platform. That's really the asset of a firm like Pixar:
their animation interfaces that let the artist draw the characters and
manipulate them.. That's why you see a given effect show up again and
again, once they've got it working.  A notable example is the computer
rendered ballroom scene from Beauty and the Beast that shows up in a Shell
Credit Card commercial, or these days, hair and clothes draping effects.
In the 90s, I think it was various particle simulations (fire and smoke,
in particular), although I remember everyone being impressed by Digital
Domain's work with the Harrier jet exhaust in True Lies in 1994. I worked
in the physical effects business then, and our stock in trade was fire,
smoke and rain, so "good" fire CGI was a threat to our business..
Fortunately, it's still cheaper to light a fire than simulate it, and no
amount of CG and green screen and acting talent can replace actual wet
rain falling on the couple saying goodbye forever in the train station,
even if the train and station are composited in later and the actors are
standing indoors with some techs hosing them down from a bucket lift above.

Lest you think this all is inexpensive, just look at the number of people
in the credits for the digital post/animation.  Animator union scale for
the highest grade is around $40/hr, lowest is around $30/hr.  You have an
army of 50 folks working on a feature for 6 months and that's about
$1.5-2M, just in wages.  The $30/hr person isn't doing high end
animation.. No, they're doing stuff like coloring or rotoscoping or match
framing. IMDB gives something like 300 people worked on Wall-E for 4
years. At a conservative 150k/workyear, the 1200 work years would be just
under $200M

The real cost is not in the rendering, it's in building the models, and
then figuring out motion paths. An old estimate I saw was something like 1
work hour to define and move a polygon/object.  I suspect it's faster now,
but then, there's a lot more objects.  In reality, features get a budget,
and then they flow that down and get as much as they can for the fixed sum
of money.  Other estimates are that an animator produces about 20
seconds/work month.  There's also a whole lot of aspects due to the lack
of scalability.  You can't get 100 animators to each do 0.1 seconds of a
character and hope that it looks right (although I'm sure the big
production houses would love to be able to do that)

One thing that has gotten VERY cheap is things like wire removal (e.g. If
someone or something is suspended from a crane, and you need to remove the
"wire" from the shot).  That used to be very expensive, and essentially
done by hand on a frame by frame  basis, but sometime in the mid 90s, Avid
(I think) came out with an engine that could do it automatically, with
minor touchups.  That made a HUGE difference in how you shot live action.
You could suspend things from 3/8" steel cables rather than thin tungsten
or piano wire.

Chroma keying (green screen) has also gotten very good because of the
computational horsepower available. It would be a real bad job if you
could see any artifacts on hair or fuzzy clothing these days.

The compositing process these days is a lot easier because they've got
good motion control and recording on the cameras, so they can match the
animation to the camera move. (and they can previsualize the moves ahead
of time, to see if it's going to "work")

On 11/6/12 6:49 AM, "Prentice Bisbal" <prentice.bisbal at rutgers.edu> wrote:

>On 11/05/2012 08:16 PM, Vincent Diepeveen wrote:
>> Jim as someone who produced games, this is not how it works for most
>> movies/animations/commercials where graphics work is needed.
>I see a big flaw in this logic here. Games !=
>movies/animations/commericals, so I don't see how 'producing' games
>makes you qualified to talk about movies/animations/commercials.

More information about the Beowulf mailing list