3D TV and 3D Cinema Tools and Processes For Creative Stereos
3D TV and 3D Cinema Tools and Processes For Creative Stereos
3D TV and 3D Cinema Tools and Processes For Creative Stereos
3D TV and 3D Cinema
Tools and Processes for Creative Stereoscopy
Bernard Mendiburu
with Yves Pupulin and Steve Schklair
11 12 13 14 5 4 3 2 1
Printed in the United States of America
Dedication
v
To frank verpillat*
“Bernard, within a few years, we’ll see the first 3D TV channels. For they won’t
be able to produce enough content for 24/7 operation, they’ll rely on content
converted into 3D. I want you to design a 2D/3D conversion workflow and
pipeline.”
frank verpillat* (October 1999)
Acknowledgments
ix
It’s obvious that the current interest in stereoscopic television derives from the
success of stereoscopic cinema, and this success can be laid at the doorstep of
three companies: Texas Instruments, with the development of their DLP (digi-
tal light processing) light modulator; the Walt Disney Company, with their
lead in the charge for 3D movies; and RealD, with its successful deployment
of the ZScreen added to the DLP projector. The basis for modern stereoscopic
projection technology, and its reason for success, is that only a single projector
needs to be employed for the successful display of both left and right images
in a field-sequential mode, as originally demonstrated by StereoGraphics for
industrial applications. This makes for superior geometric and illumination
coordination of the left and right channels, allowing for routinized operations
in the projection booth. Similarly, a virtue of today’s 3D TV is that both left
and right images can be displayed on the surface of a single monitor.
The success of current 3D TV technology, just like 3D motion picture tech-
nology, depends on viewers wearing eyewear. The eyewear that are employed
today are either passive, using polarizing lenses, or active, using electro-optical
shutters, and are identical in function to the products offered in the past. There
are advantages and disadvantages to both techniques, yet as this book is being
published, no one can predict the level of acceptance on the part of the public
for stereoscopic television that requires the use of glasses. Despite the fact that
every major television set manufacturer, as well as the manufacturers of cam-
eras, custom integrated circuits, set-top boxes, Blu-ray players, and the like, not
to mention cable companies, satellite broadcasters, Internet protocol providers,
and manufacturers of multiplexing and switching apparatus for the backbone
of the television infrastructure, despite the fact that all these players are making
a tremendous amount of effort and outlaying enormous expenditures, every-
body in the field is making a big bet on the public’s acceptance of 3D TV with
individual selection devices—that is, glasses.
There are applications today that do not necessarily require the use of eye-
wear. These applications, most probably involving laptops and handheld
devices, use interdigitated stereo pairs in a vertical columnar format with
an at-the-screen selection device, typically a lenticular screen or a raster bar-
rier. In applications where the user can carefully position himself or herself,
or by hand-holding the device to position it, and where there is only a single
user, an interdigitated stereogram makes sense. Such selection devices have
limited head-box area, so proper positioning or the relationship between the
display and the viewer’s eyes is critical. But such a stereogram is not a general-
purpose solution that must await a large number of views and a proper selec-
tion technique.
It is safe to say that if an autostereoscopic (non-eyewear) television system
could be developed and offered for sale at a reasonable price, and had image
quality equal to that of images viewed with eyewear, it would be successful.
The basic technology has been under development for about a century. One
of the several problems that must be addressed no matter what selection
Introduction xv
In this first chapter, we will try to describe as accurately as possible the landscape
in which the concept of 3D production is currently being redefined. This descrip-
tion represents only a snapshot of our professional universe as of summer 2010,
as the landscape itself is an ever-changing picture. Still, we feel it is important to
set the background before jumping into the story. Describing how new tools are
being invented every day will help the reader manage any new task that may be
handed out to 3D crew members.
3D movie analysis
80
70
60
50
40 5
1 69
7
30
3
6 48
20 7
5
3
10 7
8 9
1 8
3 4
– 1
2008 2009 2010 2010 2011 2012+
(Released) (Announced)
Native 3D CGI 2D 3D conversion Unknown
Source: www.3dmovielist.com, PwC analysis. Note: The list includes short movies.
FIGURE 1.1
3D movie releases 2008 to 2012.
Image courtesy of Vincent Teulade, PricewaterhouseCoopers Advisory.
3D TV Is Born in 2010
Now well established in cinema, stereoscopic content will soon invade your
living room with 3D TVs showing 3D games, TV channels, and 3D video on
demand (VoD). TV is now exactly where cinema was a few years ago in terms
of technologies and economic background. For anyone following the 3D revo-
lution, the only thing surprising about the 3D TV deployment is its speed. TV
is at the edge of a major shift, and “going HD” is only the tip of the 3D iceberg.
Steve Schklair, one of the forerunners among 3D TV advocates and engineers,
recalls the key events leading to 3D at home. Readers willing to learn more
about the inception of the entire 3D production system and procedures will
find Steve’s extensive recapitulations of the early years in Chapter 6.
STEVE SCHKLAIR
In April 2008, 3ality Digital did the first 3D broadcast, with Howie Mandel doing a
comedy skit from our offices in Burbank to the Content Theater at NAB as part of my
speech. This is the event that kicked off the BSkyB (British Sky Broadcasting Group)
launch of the world’s first 3D channel by a major broadcaster in 2009.
In September 2008, 3ality broadcast Jeffrey Katzenberg live to the International
Broadcasting Convention (IBC) in Amsterdam as part of my keynote speech. This was
the first transatlantic broadcast of 3D.
To give you a more general point of view of the 3D industry, we use a classic
three-act structure to represent the rapid transition to 3D TV in the few months
between 2009 and 2010.
4 3D TV and 3D Cinema
Users do not buy a technology, they buy the experience that comes from it.
Sony announced that they would produce the 2010 FIFA World Cup in 3D and
ESPN would broadcast the 25 matches on their new 3D TV channel. Hours
after that announcement, “ESPN 3D” was ranked as the number five search on
Google. The audience’s craving for 3D was established.
CES 2010 also saw the launch of the 3D extension of the Blu-ray format and the
presentation of Panasonic’s 3D camcorder, the soon-to-be-famous AG-3DA1.
Main Critics of 3D
In order to give a fair assessment of the current perception of the 3D trend in the
general audience, we have to mention the 3D-bashing hype that could be found
on the Internet in the summer of 2010. It came from a variety of sources, from
inexperienced bloggers to seasoned and respected media journalists. This back-
lash was to be expected, both as a natural response to the marketing pressure and
as a consequence of new 3D technology shortcomings or inappropriate use by
some opportunists. Let’s have a look at some key elements mentioned during the
summer of 3D bashing and explore whether there is any value in these criticisms.
with passive glasses. Comparing the two is kind of like comparing apples and
oranges: they have very different purposes, specifications, and price points.
The first generation of active 3D glasses was vendor-specific, but they can be
made compatible with different TVs after some light polarization and synchro-
nization protocol adjustments. However, these adjustments confused reviewers
to such an extent that some saw in the glasses the start of a new format war—
one based on display technology rather than content distribution.
In early January 2011, as we finalized this book, we introduced the world’s
first hybrid 3D glasses, which work in both active and passive modes, at the
2011 Consumers Electronics Show in Las Vegas. This new product, designed by
Volfoni and called ActivEyes, is for both professional and personal use, allow-
ing directors in OB vans to use passive monitors for camera checks and active
3D TVs for beauty and program monitoring at once.
Here is another lesson for the CE industry: The HD optical support format war that
left the HD-DVD dead and the Blu-ray gasping for a 3D defibrillation, actually did
hurt the audience confidence in home cinema technologies more than we thought.
develop and five years to produce. This is a prohibitive amount of time when it
comes to television technologies and storytelling techniques.
In television production, the longest timeframe is one year—and in most cases,
it is only a few months between the invention of a storytelling tool and its
common use in talk shows and sitcoms. In news, it is weeks, or even days.
As soon as music video TV channels are 3D, stereoscopic creativity will get its
ADHD treatment. As soon as nature and animal TV channels are 3D, the “we
take you there” shows mixing detailed wildlife close-ups and gorgeous vistas
will have their steroid shots. If you want to have the faintest idea of how far
3D sports coverage will evolve in 2011, just have a look at the first and last FIFA
3D matches produced in 2010. Live stereoscopic production will grow up right
before our eyes, as we witness camera operators and directors learning and
expanding the tools of the trade while producing 2011 sport events in 3D TV.
As a teaser, though, let’s have a quick look at the conditions in which such pro-
ductions occur, that is, how 3D gets involved in these branches of the enter-
tainment industry, and how they are affected by it. As a 3D TV executive once
summed it up: live television brings the event to you, HD quality seats you in
the stadium, and 3D realism throws you onto the field. How does this affect
content production?
3D Broadcasts
A recurrent project for many a broadcast research centers was 3D TV produc-
tion. We have now entered a new era where 3D TV production has escaped to
the entrepreneurial world, which has started to self-finance its projects. Take a
look at the following timeline:
It is now possible to put on a live 3D broadcast that equals the quality of what
was post-produced only a few years ago. As an example of this progress, refer
to Figure 1.3, which shows the production of a Black Eyed Peas concert in 2010
compared to the U23D feature concert shot in 2007.
Professional 3D Production in the 2010s Chapter 1 11
Alternative 3D Content
While watching 3D mature and experiencing the major production paradigm
shift currently happening, the experts in the know were on the lookout for the
“CinemaScope” of the 2000s. What was the new media technology that would
kill 3D, just as widescreen viewing ended the golden age of 3D in the 1950s?
Alternative content (AC), like live sports or musical events, was seen as the
most serious contender, with dynamic seats and multiplayer games far in the
distance. As it happened, the 3D wave was so strong that AC itself had to go
3D, as you see now with 3D sports broadcasts in theaters.
Our thesis is that 3D will remain the strongest effect of visual media digiti-
zation, even if AC in theaters is another side effect of digitization. In reality,
3D and AC are complementary: the first generation of 3D broadcasts, such
as FIFA 3D, have to cope with 3D TV screen scarcity, but by also serving as
AC for 3D theaters, they balance their budgets. The current low rate of 3D
TV equipment acquisition can make it worthwhile for theaters to show AC
events while they are broadcast on 3D TV, and thus host the biggest 3D TV
party in town.
3D Documentaries
Stereoscopic displays are used in virtual reality for the enhanced realism and
added navigational skills they provide. The purpose of a documentary is to
present a vision of reality to its audience—so, with its realistic effect, 3D will
soon prove itself to be an excellent tool for that specialized movie genre. Will
this effect be the same for social documentaries and nature movies?
It’s no wonder that Discovery was the second major U.S. TV franchise to publi-
cally commit to a 3D channel (after ESPN). This involvement in new media
sparked a whole set of independent 3D documentary productions in the sum-
mer of 2010. For better or worse, producers and DPs with no 3D experience are
jumping on the boat with low-end 3D rigs.
However, documentary production is not particularly suitable for complex or
fragile shooting equipment; its budget does not allow for renting high-end rigs
for weeks in the bush, and the shooting environment is often harsh for surface
mirrors and micron-accurate motion controls. If there is a branch of 3D pro-
duction looking for the monopack of 3D, it is the documentary.
Professional 3D Production in the 2010s Chapter 1 13
3D Feature Films
Since 2005, the number of digital 3D films already released, slated for release,
or green-lighted has reached the number of 3D films actually released in the
1950s. Even if the current 3D era were to fade out, it would have generated the
same quantity of content as the 1950s golden age. Still, the comparison may
end there. First, that golden age lasted only two seasons, 1953 and 1954; and
second, the types of movies produced were totally different.
In 2010 we see a relatively varied mix of blockbusters converted to 3D, low-
budget movies shot in 3D, and a flock of CGI animation rendered in 3D. After
Avatar, we have yet to see another major live action blockbuster shot in 3D
rather than converted in post, and this should be the case in 2011 as well.
As 3D champion Ray Zone said two years ago: “The train has left the station,
the genie is out of the bottle; cinema is going 3D. Where it will go now is not
anymore a technologist question. It’s a business and art issue.”
Despite the 3D screen count continuing to increase by good numbers, feature
movies keep facing severe 3D screen scarcity. In 2010, we saw two 3D mov-
ies from the same animation studio, Shrek Forever After and How to Train Your
Dragon, fighting for 3D screens. The year 2011 will be no different, with a new
3D film coming out every two weeks.
aspects: it can be copied infinitely at a marginal cost, and it can be compressed thousands
of times and keep its visual quality. Infinite copy generations gave us compositing, the
most modern visual effects, and virtual editing. Compression gave us high-definition TV
and illegal downloading, otherwise both of which would be far too expensive to operate and
may not even exist. There’s no spare bandwidth to be lost in broadcasting analog HD.
Digital technologies had an impact on content production. With CGI images that make
any visual effect or stunt digitally possible, they extended the domain of suspension of
disbelief and led us into a world where we just couldn’t believe our eyes. This affected
story and image realism, but it did not affect the visual art in the same way as motion
or sound, or even color, did. Digital 2D brought us minor changes in visual storytelling.
HD provided increased resolution, but it did not change the audience experience. As for
the audience’s sensory experience, the gain from over-the-air analog TV to digital TV is
much more noticeable than from standard definition (SD) to HD. Similarly, there’s much
more of a difference between VHS and DVD than DVD and Blu-ray. If digitization changed
our visual world to some extent, HD did not change the visual rules; artistically, there
was no HD. Many HD production trucks still have 4/3 markers on their preview monitors.
Technologically, 3D is a sibling of HD, an offspring of digitization. If it were not for the
digital revolution, there would be no 3D; it wasn’t until it was possible to tweak images
at the subpixel level that 3D postproduction took off in the 1990s. Stereoscopic
movies escaped from high-end large-format special venues when digital projectors
became available, and that 3D mode was their best sales pitch. Once again: there
would have been no digital projection outreach without 3D. The 3D renaissance and
digital revolution cannot be separated.
Now consider how, in many cases, the upgrade from HD to 3D is marginal in cost
and complexity. Thanks to the computerization of CE and production gear, 3D is
often a software upgrade on HD. When it comes to hardware, HD and 3D are twin
technologies. Many HD products can be retrofitted into 3D with a simple software
update (Sony PlayStation 3 and some Blu-ray Disc players, video switchers and mixers,
image encoders, and transport equipment) or a minor electronic upgrade like 120 Hz
TVs. Some HD products are 3D-ready with not a single knob to turn on—like RPTVs
(rear projection TVs) and DLP projectors or SRW video tape decks.
Technologically, there was no HD; it was just that 3D software was not yet ready when
digital hardware was deployed.
Thus, there’s a difference between 3D and HD: 3D is changing the audience experience,
and 3D will eventually change the way we make movies and the way we produce live
sports on TV—like color or electronic news gathering (ENG) did in their time. It’s not an
issue of technology and skills; it is about art and direction, like camera positions and
editorial pacing. HD did not impact the visual arts as much as 3D will. Artistically, there
was no HD, but 3D is waiting to hatch.
The original cause of the revolution is digitization, and the end result is 3D. HD was
a transitional byproduct, which barely lived for a decade, almost never found its
packaged media, and did not influence the art.
Still, thanks are due to HD for paving the way for 3D and taking care of the upgrade
costs from analog, SD, and 2D television.
Professional 3D Production in the 2010s Chapter 1 15
Michael Bergeron provided reference numbers that he found in the research literature
for the maximum sustainable retinal rivalry. You’ll obviously want to stay as far away
as possible from these numbers in your productions, for they set the bar for half your
audience leaving the room in tears.
Maximum Sustainable Retinal Rivalry:
n Vertical .8%
n Rotation .8%
n Zoom 1.2%
If you bring some visual effects or graphics into your picture, your accuracy
requirements get much higher. If you want to credibly match stereoscopic con-
tent and CG images, you’ll need to match practical and virtual camera angles
much more accurately than in 2D. This means that if your 3D footage is not
“clean,” you’ll have to ask the FX team to match its roughness.
Current 3D productions will most likely be shown on large screens, as cinema,
or as alternative content. Planning for 3D TV releases may not make 3D prod-
ucts any easier to produce, as shorter viewing distance and higher light level
make 3D TV imagery at least as demanding as 3D cinema.
sets up the stage. TV production’s working schedules and procedures have been
finessed and optimized for years, and 3D can hardly change anything there.
You just can’t be slower than 2D. The real world is not waiting for 3D; it’s 3D
that has to come up to speed.
You will not change that fact in the movie industry either; your 3D camera
units must be able to be set up just as fast as 2D. A film crew moves from set to
set, from location to location, and shoots as many takes as possible in a single
day, so fast setup is even more important in feature films.
Live 3D: You Won’t Shoot It Again, You Won’t Fix It in Post
With staged 3D, there’s no chance to shoot it again. The next president of the
United States is not likely to take his pledge twice, the race winner will not
cross the finish line again, and that gnu will not resuscitate just to be killed
again, even if the lion seems to have had a lot of fun. Live shooting is not only
about being ready when it happens; it is also about shooting it the right way,
right away. Fixing the convergence in post may be an option, but fixing the
inter-optical (IO) distance is not. Just as you can’t save an out-of-focus shot,
fixing shooting accidents in post is too expensive in 3D. You’ll never get a bet-
ter deal than shooting right.
Of course, in a live 3D broadcast, you won’t even have a chance to fix it in post.
If it was wrong, it was broadcast anyway, and did hurt the audience. There are
many 2D imperfections you can get away with. Not in 3D.
Professional 3D Production in the 2010s Chapter 1 17
STEVE SCHKLAIR
When using image-domain geometry correction in a live production, all camera feeds
need to be fixed before they enter the production, not just during the program and
preview.
Frame rate is a crucial issue in 3D, and its standards will have to be increased in TV and
cinema in the next 5 to 10 years. Currently, we’re still struggling with progressive and inter-
laced standards in 3D, but soon we’ll have to face the fact that the brain is more sensible
to stroboscopic effects in 3D than in 2D. The deep depth of field that some stereogra-
phers recommend for 3D doesn’t help either. As such, a lot of sports content would bene-
fit from being produced in 720 at 50p or 60p; right now it is most likely produced in 1080
at 50i and 60i. This is an issue with the current generations of 3D displays that combine
the two fields and double flash them at 120 Hz. Instead, the proper process should be to
interpolate each field to full resolution separately and flash them once each.
There’s a closely related issue with the interlaced-to-progressive converters used in
3D: they should actually generate four images, not just two. But solving this requires a
significant redesign of the workflow and tools.
Professional 3D Production in the 2010s Chapter 1 19
STEVE SCHKLAIR
The story is the most important element. Gratuitous wow effects do not move a story
along. They do the opposite, and take an audience out of the story. These are shots
that are screaming out, “Look at me! Look at me! I’m 3D!” Still, every audience would
like a few wow 3D shots, with negative parallax. These have to be carefully scripted
in, if used at all. Perhaps the titles in a feature or the graphics in a broadcast are
enough. Think of the early days of surround sound, where every soundtrack swirled
the sounds around the room in a dizzying fashion. Over time this settled down, and
now the surrounds are used creatively, or just to create atmosphere. Eventually, and
we hope very soon, 3D visuals will settle down to a mature level and exist solely to
add to the filmmaker’s craft of storytelling.
BRUCE AUSTIN
Shooting 3D in Africa is like putting your rig in a washing machine, running a full high-
stain cycle, and then checking to see if there’s a missing screw.
Despite all its limitations, its long focal length being one, I love the Panasonic
AG-3DA1. In real nature documentary situations, rigs just do not work; we have to get
rid of them at some point. They are too dust-prone, full of chromatic and geometric
aberrations, and heavy and fragile to boot. You feel like you’re hurdling around a studio
camera in the bush, as though someone in the camera department played a prank on
you when packing your stuff. We need cameras 5 mm in width to make practical SBS
(side-by-side) rigs for animal documentaries.
20 3D TV and 3D Cinema
If you need 15 seconds to set up and 10 seconds to set your IO and convergence, you
have to remember that, on average, an animal leaves within 60 seconds. I eventually
learned to set up the camera while we were far away from the scene, guesstimating
the 3D, and then we would roll up to position. We had to leave the engine on, other-
wise the animal would leave, and that created vibrations.
FIGURE 1.5
Producing Live 3D Reveals Unexpected Effects. A still camera flash generates a halo of focused
imagery in the middle of a tracking blur.
Side-effect catch by Nick Brown, Digital 3ality, during a BSkyB 3D football production.
expectation suggested that the smaller screen size would make the 3D imperfec-
tions less noticeable; vertical disparities of a few pixels would be troublesomely
big on a theater screen but unnoticed on a TV set. However, experience shows
that this is not the case. Is it due to the sharpness and luminance of the image?
It’s not yet clear, but the lesson of the day is that 3D TV is not cheap 3D.
In addition, the distance between the viewer and the screen reduces the screen
size effect. As a result, most of the depth settings are relatively constant, which
means it’s primarily the depth placement that evolves. Producing for large
screens and small screens still dictates different framing and editing—but the
differences are not the ones that were expected.
Fortunately, the 3D industry seems to get it; we are beginning to see more and
more collaborative work, especially in the realm of education.
The Stereographer
The stereographer used to be the jack-of-all-trades of 3D productions, bring-
ing stereoscopic expertise to the project, from development to projection. With
3D production increasing in size, duration, and complexity, the job tends to
be shared among a principal photography stereographer, usually assumed by
someone with a DP or AC background, and a postproduction stereographer,
usually someone with an FX supervisor or compositing background.
In a live TV production, the stereographer is more of an instructor who will
teach the ENG team how to use the 3D gear and supervise the pool of conver-
gence pullers or 3D image analyzers that control the depth in real time.
On the movie set, the stereographer sets the depth, sometimes in collaboration
with the DP or director, sometimes independently. This will eventually change,
with the stereographer role gradually fading out. Setting the depth will then
be another artistic tool utilized by the DP—just as it is already the DP who
is responsible for the continuity of depth of field, or using depth of field to
enhance storytelling (working closely with the director, of course, as is already
done). There may or may not be a 3D-AC in charge of actually manipulating
the rig.
In 3D Movie Making, I signaled the paradox that required a stereographer to
have experience in 3D and in feature films even though no 3D feature films had
been produced for many years. That paradox has now evolved into a new one:
Good stereographers are those who will eventually make themselves useless,
because in every 3D movie or 3D broadcast, they teach the camera operators,
ACs, DPs, and directors how to make good 3D. Eventually, this 3D knowledge
24 3D TV and 3D Cinema
will be pervasive in movie crews; everyone will have some general knowledge
of 3D, and will also master the specialized skills needed for his or her job posi-
tion. By then, a stereographer may or may not be useful on set—in the same
way that there are no longer color technicians. On the other hand, there are
still sound engineers. Time will tell which way the road goes.
It is also possible that all stereographers will eventually be stereo supervisors or
VFX supervisors. If 3D is a new visual storytelling tool, then it is the DP who will
be responsible for using this tool to tell the story. It is worth noting that the gen-
eration of stereographers who worked on the film-era 3D movies had a camera
background, while the digital generation comes from visual effects supervision.
Knowing how much 3D is an end-to-end quality issue, more and more stereog-
raphers are reluctant to get a job where they do not control 3D quality from
shoot to post. As stereographer Eric Deren says, “Someone in post can very eas-
ily mess up what you shot, as soon as you are gone. Even worse, you may have
to deal in post with the nightmare of fixing dismal 3D that somebody else
shot. In both cases, the end result is poor or bad 3D that is reaching the screen
with your name on the credits and your reputation at stake. Whether they are
the DP or the stereographer, people shouldn’t be paid until the post is done.”
27
The type of camera or rig you’ll need for filming 3D is dictated by various FIGURE 2.1A
parameters—and for quite a long time, the number one rule was availability. Compact 3D Camera.
The JVC GSTD1BUS is a
However, this has changed since 3D entered the realm of mainstream enter-
3D camcorder designed for
tainment, with the recent increases in vendors of 3D tools, the expanded avail- run-and-gun productions.
ability of 3D gear at rental houses, and the
global growth of 3D production budgets.
1. All-in-one 3D camcorders.
2. Small 3D rigs, for a steady cam or shoulder
mount, using digital camera heads.
3. Large 3D rigs, hosting full-size digital cinemas,
broadcast cameras, or camcorders.
Now that we have made the case for 3D rigs, let’s acknowledge how complex the
system is, and explore how and why it came to that level of complexity. After all,
a 3D camera is just a pair of 2D cameras stitched together. Get a wood plank,
two screws, and some duct tape, and there you have it. To be honest, that’s pretty
much how most stereographers started in their younger years. The richer had alu-
minum plates and the smarter got sliders, but we all started with the basics. And
the few of us who pursued the journey all the way to the current state-of-the-art
3D rigs ended up with the same solution: a big bulky camera assembly revolving
around a half-mirror, hosting a dreaded number of motion control magic boxes,
and most likely connected to an image processing behemoth. How did we all
end up at the same point, each one following our very own path? Let’s fast-
forward 20 years in 3D camera design to understand.
First, there were the film camera rigs designed for theme park movies in the
1980s. Sometimes carved out of a single piece of aluminum, they held together
two cameras and the half-mirror splitting the light beam. They were beauti-
ful and accurate, but heavy, expensive, and not that versatile, with no dynamic
inter-optical distance.
Change came in the 1990s, with the new wave of electronic imaging announc-
ing the digital revolution. Camera heads could be detached from processing
and recording units, and motion control started to demonstrate its capacities.
At that point, a few engineers or cinematographers on the planet established
the 3D equation:
Electronic imaging motion control 5 perfect 3D images
They started developing what would eventually be the blueprints for the 3D rigs
you see today. That design was developed the hard way, by getting back from
location late and often exhausted, with less than optimal pictures, and cursing
about not having the 3D tools that would make it easier. Remember, produc-
ing 3D always starts by putting two cameras together and shooting. In that pro-
cess one discovers the limitations of the system, builds solutions to circumvent
them, and progresses toward higher 3D quality, easier production schedules,
and more resilience to deal with the unexpected issues that content production
always randomly, yet steadily, generates.
If you had been making 3D movies, and therefore 3D cameras, for the last
20 years, here is what you would have discovered:
Step one: You need to motion control the left and right camera’s relative
movement to generate good 3D images, because
1. Cameras need perfect alignment, requiring XYZ-axis translation and
rotations.
2. One needs to be able to set up the inter-optical distance and convergence
almost instantly.
3. On shots where the camera or the action is moving, inter-optical and
convergence has to be dynamic.
30 3D TV and 3D Cinema
Step two: You realize that your inter-optical is too wide. In most movies shot
in 2010, inter-optical is under half the human inter-pupillary distance, which
is about three centimeters. This means that the cameras need to be closer than
the lenses’ width:
1. Cameras need to shoot through a 45-degrees half-mirror.
2. The mirror has to be a surface mirror.
3. Mirrors are extremely prone to getting dusty.
4. A mirror is not a perfect plane; it bends under its own weight.
5. Surface mirrors have chromatic asymmetry.
6. Most importantly, the mirror’s surface is fragile, and hitting it with a lens
can scratch it.
Step three: You need to use zoom lenses, because changing lenses is a complex
task on a mirror rig. Hopefully, you are already in the late 1990s, and the new
HD zooms are sharp enough to be used as adjustable primes. When zooming
in with your 3D rig, you discover that
1. Zoom progression is not linear and changes from lens to lens.
2. The optical axis shifts along a progression, and that shift is different with
every lens you try.
3. As a result, each lens pair has specific alignment requirements that vary
with focal length.
Step four: You have now reached the early 2000s, and computing power is
soon to become a commodity. You safely assume you can build zoom look-up-
tables (Z-LUTs) that use the XYZ motion controls to align the camera’s optical
axis. Here comes a new batch of bad news, however:
1. Manually generating the Z-LUT is a time-consuming process.
2. You should manually generate the Z-LUT every time a rig is assembled.
That last point should have you raising your eyebrow. What does this mean?
It means that if you take the very same set of equipment and rebuild the very
same rig, you won’t end up with the very same results. First, assuming you can
perfectly repeat procedures in movie and TV production is not such a safe bet.
Second, we need to talk numbers here. You want to get to less than a couple
of pixels of vertical disparity on your final image. When the sensor has 1080
lines of photosites on a 1/3-inch sensor, one line is 1/1080th of 3.6 mm, close
to three microns. The larger digital sensor, the Phase One P65, with 6732
lines on 40 mm, has a whopping 6 microns per pixel. To give you a compari-
son base, a human hair is 17 to 180 microns. You cannot build a mechani-
cal assembly with moving parts and expect alignment figures to be predictable
within a few microns.
1. The order of magnitude is to move camera and lenses weighing 10 Kg at up
to 5 cm/sec.
2. Inertia and momentum effects set a physical limit that your hardware magi-
cians will not be able to break.
The Stereoscopic Camera Unit Chapter 2 31
You still need an automatic lenses matching tool, and you build your image
analysis tools:
1. Automatic Z-LUT creation is a quite complex image processing problem
involving coplanar geometry.
2. Over a typical shooting session, thermal dilatation can change a Z-LUT or a
camera geometry. Just touching a camera may be enough to change the cor-
rection values.
Step five: You have now successfully reached the late 2000s, and you are the
inventor of one of the few world-class 3D camera systems. By now, it is obvious
that 3D cinema is going mainstream, and 3D TV is right around the corner. Both
require that you produce perfect 3D imagery, and both won’t cope for long with
the idea that every 3D shot is an FX shot that should be groomed in post.
1. Manual depth setting by human operators is not adapted to 3D TV
productions.
2. Full frame analysis and correction should be performed in less than 40 ms
for live TV.
3. All units involved in a multicamera production should be depth-coordinated.
After reading this list, you may wonder if it’s worth it—perhaps we are overdo-
ing it, and the answer may be to start from a simpler place, or to get more “fix
it in post” power.
Zoom Lens
Zoom Lens Motorization
Mirror
Zoom Lens
Motorized Interaxial
and Convergence
Adjustment
FIGURE 2.2
The Main Parts of a
3D Rig.
The yaw control is used to compensate for the zoom axis mismatch. Note that
this mismatch is in both X and Y directions, so it will affect the convergence angle.
As a result, the convergence control is actually used to correct the zoom axes.
When discussing mechanical controls, it makes sense to mention the rolling
shutter and camera orientation issue. Most first-generation rigs had the vertical
camera with its top facing the set. This meant that if you looked at it through the
mirror, it seemed to be upside-down; indeed, that’s the way it looked at the scene:
upside-down. If your cameras had a rolling shutter, as most CCDs do, it meant
that they scanned the image the same way video cameras used to, or the same
way a scanner or a Xerox copier does. In the 720p U.S. video standard, short
for 720@30p, the norm is to scan the image top to bottom in a little less than
The Stereoscopic Camera Unit Chapter 2 33
FIGURE 2.4
ARRI Alexa on a 3D Rig. Note the threaded holes on the sides of the rig, used to fix accessories or
to bolt the rig on ad hoc mounts.
The Stereoscopic Camera Unit Chapter 2 35
to record
n Back position shot, framing the whole scene
for a concert
n All shots using a miniature point-of-view
FIGURE 2.6
Side-by-Side Cameras
for Helicopter Shot.
Helicopter setup in
the Makay Desert,
Madagascar.
Image courtesy of Binocle/
Gedeon.
38 3D TV and 3D Cinema
THE HALF-MIRROR
The mirror in your 3D rig is not just any mirror: it’s a 50/50 surface mirror.
This means that exactly half of the light is reflected (reflected light) when the
other half is let through (transmitted light). Reflected light and transmitted
light, define the transmissive camera and reflective camera. The mirror has to
be a surface mirror, it means that its reflective coating is on top of the glass,
not under it. Walk to your bathroom mirror and put your finger on it; you’ll
see the thickness of the glass between your finger and its reflection (actually,
you’ll see twice the glass thickness). Now go to the kitchen and grab a brand-
new, never-been-scratched stainless steel pot—one you can see yourself in—
and put your finger on that. There is no space between your finger and its
image. This is because the pot is a surface mirror, and your bathroom mirror
is not.
To understand why the 3D mirror has to be a surface mirror, try another
hands-on experiment. Get that bathroom mirror to your office video projector
and use it to divert the light beam. You’ll see a duplicated image, with a sec-
ondary picture that is approximately 10% the brightness of the main one. This
is because the glass surface of the mirror is not 100% transparent, and thus it
is a mirror in itself. If your rig were not fitted with a front surface mirror, the
reflective camera would see that double image.
n Inadequate mounting apparatuses, which causes n With the help of an autocollimator, check the colli-
distortion of the substrate and severe degradation mation of both reflected and transmitted cameras
of the reflected images. through the beam-splitter.
n Test your rig in the horizontal and tilted down posi-
It’s also important to know how to test a beam-splitter:
tions to make sure the beam-splitter is properly
n Check if the beam-splitter is mounted right side up secured and the focus and optical alignment is not
by bringing the tip of a pencil to the mirror surface. affected when the rig is tilted downwards.
If you see a double reflection of the pencil tip, the n Check if the beam-splitter and its mounting appara-
beam-splitter is upside down. If you see a single tus can accommodate filtration in front of the lens;
reflection of the pencil tip, the beam splitter is right you will need it to cut infrared contamination and to
side up. cut light levels when using the latest generation of
n Check for scratches and defects in the coating. digital cameras.
n Test the difference in focus setting between n Check if the mounting apparatus can accommodate
the reflected and transmitted image at wide open a clear cover to protect the beam-splitter during
iris. setup and during special effects. This will extend the
n Test the focus performance at wide open iris with useful life of the beam-splitter and save you a signifi-
your longest lens, at a focusing distance greater cant amount of money.
than 60 feet.
The average selling price for a good quality beam-
n With the help of a vectorscope and a color chart,
splitter is about US$5,000, depending on size and
check the color and flare disparities between the
quantity. However, a high price does not guarantee a
two images, and compare them against the same
good beam-splitter.
target without a beam-splitter. Perform this test
under both tungsten and daylight illumination. An Happy testing!
overlay histogram will help you visualize the color
disparities. Sébastien Laffoux, Laffoux Solutions, www.tangohead.com
FILTERS
If your mirror is not doing a perfect job of splitting light in half, one camera
will get more illumination than the other, and the apertures and depth of field
will not match. This must be fixed, for eye rivalry in sharpness, details, and tex-
ture density is prone to generate visual processing overload and mental fatigue.
This is because, first, your brain needs to match two uncorrelated images; and
second, it then has to map the finer texture onto your 3D vision, rather than
just using whatever each eye sees.
Light levels need to be balanced with neutral density (ND) filters. Stereographer
Eric Deren suggests getting some 1/8th to 1/4th stop ND filters that may be on
special order at your camera shop. If you cannot get them, use the gain setting
to balance the cameras, not the iris. Experience shows that noise discrepancy is
less noticeable than depth-of-field mismatch.
Light polarization is an issue too. A reflective surface will polarize the light in
a direction defined by the product of the surface’s normal and incoming light
vectors. A light source beaming onto a glossy wall will bounce polarized. When
it reaches the half mirror, it will go through it unaffected, and meanwhile
The Stereoscopic Camera Unit Chapter 2 41
will be filtered out from the bouncing light path. As a result the transmissive
camera will receive more light than the reflective one. The disparity is neither
persistent nor uniform with camera movement, and can amount to up to two
or three full stops in specific conditions. There are two ways to take care of it:
linear polarization and circular polarization.
The linear polarization solution places a linear filter in front of the transmis-
sive camera to replicate the light-filtering that the reflective camera experi-
ences. This filter will typically generate a full stop loss, and requires another
one placed on the reflective camera to even out the light levels. This solution
requires that you put filters inside the mirror box, and consumes one stop on
both cameras.
Facing the challenge of a night shot with polarizing surfaces like car hoods,
and no stop to lose, Eric Deren came up with a simpler setup using a quarter-
wave plate, also known as a circularizing filter. By applying this filter in front of
the mirror box, the linearly polarized light gets circularized and is much less
affected by the half-mirror. Despite being theoretically less perfect than the
two linear filters, this method has proven very effective for day shots and much
more productive.
One more word from Eric:
Fix all this filtering gear on the rig, not on the camera; otherwise, any
adjustment is likely to misalign them. Remember that the mere use
of touchscreen functions on prosumer cameras is enough to mess up
your alignment. While you’re at it, make sure the lenses are tight. Not
perfectly locking them may not even be an issue on a lucky 2D shoot
day, because they’ll stay stable enough for shooting—but in 3D it’s a
no-no not to double-check that the lenses are perfectly in position and
locked down before aligning the rig. Otherwise they’ll slip during the
day, and by the time you tighten them up, you’ll need a full realignment
of the rig.
Remember to have one replacement mirror, at the very least, because they are
amazingly fragile and hard to find once on location. Even if you can find a
local spare one, you take the chance of getting a lower quality device. Even
a comparable one might have different optical characteristics that will force
you to reconsider all your fine-tuned light filtering. Put them in a safe box,
protected from dust and hard drops—and, while you’re at it, put a handful
of anaglyph glasses in this save-the-day bag. They’ll make themselves handy
someday.
the camera in virtually every position. They are typically used for close-up
subjective shots—such as shooting dynamic point-of-view from a charac-
ter who is either a murderer or a victim of a murderer—which means they
need to have a small interaxial, or, in other words, a mirror rig. They need to
be quite stiff to hold cameras together while the operator climbs the stairs,
and they need to do all this within a very restricted weight limit that also
includes the two cameras, batteries, recorders, and 3D image processor. In
addition, you need a rig where the left and right cameras move symmetri-
cally to keep the whole mount balanced if the inter-optical changes along a
dynamic shot.
CAMERAS IN 3D
A very broad range of cameras is used in 3D productions. You’ll find that many
2D cameras are suitable for use in 3D rigs, from the consumer camcorder to
the very best digital cinema camera. That said, not all cameras will be conve-
nient for 3D. We’ll see what a 2D camera needs to fit into a 3D unit.
If there’s no rig and camera combination that meets your needs, you may want
to have a look at 3D camcorders, or, as a last resort, at 3D lenses that mount
on 2D cameras (if you have tight form factor constraints).
The Stereoscopic Camera Unit Chapter 2 43
MECHANICAL DESIGN
After genlock, the second most important requirement in filming 3D is the
mechanical design. You want to be able to fit your camera on your 3D rig. The
two key factors are a stiff tripod socket and no protruding microphone. Plastic
body cameras tend not to keep alignment when touched, moved, or simply
exposed to heat. A long-neck hypercardioid microphone will hit the mirror,
especially in reversed mount configuration for rolling shutter cameras. Keeping
the camera away from the mirror forces you to shoot with a small field of view.
As an example, the Sony EX3—despite being a great 3D camcorder with a com-
pact design and HD genlock—suffers from its ergonomic design for shoulder
operation, which makes it extremely prone to move on the rig and lose its
proper lens alignment. Connectors on the back are preferable, for any connec-
tor on the side will generate unwanted rotations that are more complex to con-
trol practically or post-correct than vertical shifts.
DIGITAL OUT
The third most important feature is a digital out. It is a must for 3D monitor-
ing, and, if needed, recording uncompressed or high bitrate on external DDR
(Digital Disk Recorder). SDI out will be the best, with HDMI as an interesting
low-cost contender. Many 3D multiplexer boxes now offer HDMI inputs. It is
important to understand that until you can get a live full resolution image out
of the camera, you do not have a 3D camera. You do not really get into 3D
production until you combine the two image sources in one device, whether
44 3D TV and 3D Cinema
FIGURE 2.8
Rolling Shutter Affects Image Symmetry. In this traveling shot, the image is skewed in opposite
directions by the combination of the movement and the scan. When displayed in 3D, the asymmetrical
skew generates false depth and geometries, rotating the whole scene toward the audience.
The Stereoscopic Camera Unit Chapter 2 45
ARRI ALEXA
http://www.siliconimaging.com/ The Alexa is the latest digital cinema camera
DigitalCinema/ from ARRI, a renowned German company. It is
being used by many cinematographers on 3D
projects, including Martin Scorsese on The Invention
of Hugo Cabret. The clock and timing subsystem was especially
tuned for multiple cameras. Because this camera is still a very recent addition
to the cinematographer’s toolset, you will have to look for more information
on the book’s companion website.
The Stereoscopic Camera Unit Chapter 2 47
FIGURE 2.9
SI-3D Rig with Cine
deck. On this small 3D
camera unit, images
coming from the SI-2K
are recorded on the
Cinedeck SSD.
Image courtesy of Element
Technica.
FIGURE 2.10
The ARRI Alexa. Two
ARRI Alexas on a Stereo
Tango rig.
Image courtesy of
Sébastien Lafoux.
SONY HDCP1
The Sony HDCP1 is a compact point-of-view camera designed to fit into a 3D
rig. Its 68 mm narrow form factor is conveniently close to the 65 mm average
human interaxial distance. Its top and bottom tripod socket allows this rolling
shutter camera to be reverse-mounted as a reflective camera. All the connectors
48 3D TV and 3D Cinema
and memory card accesses are on the back of the body, and the lens can be con-
trolled through the camera that passes the commands from the SDI to the lens
control connectors. The system is currently extended to send commands to the
rig’s motors using the Element Technica protocol that has been made public.
3D Camcorders
Even in 2D, some productions will not use a full-size camera. Some can’t
afford it, and some don’t need it—for whatever reason, they require smaller or
cheaper solutions. Even on large budgets, you may want to shoot some B roll,
or have a camera you can play with without endangering your carrier if some-
thing unfortunate happens. Why would it be different in 3D?
Compact, single-body, all-in-one 3D camcorders have been the holy grail of
stereoscopic principal photography for quite a long time. Unfortunately, the
more they approach the real world, the better we see their limitations.
The advantages of an integrated 3D camera are obvious. One power switch,
one record button, one stereoscopic file. No cables, no mirrors, no screw-
driver, no surprises. The drawbacks come with the fact that 3D photography
is all about the relationship between the set and the camera metrics. The usual
process is to have a scene, and to adapt the camera to it. Recent statistics have
shown that most 3D shots, in CGI or live action, have an inter-optical ranging
between 5 and 20 mm. All-in-one cameras have either a fixed inter-optical at
70 mm, or a minimum inter-optical larger than 30 mm. As a result, short of
adapting the camera to the shot, you’ll have to adapt the shot to the camera.
You’ll want to stay away from the first plane objects, keep the background close
enough, or both.
Still, it’s easier to stay far away from that lion killing an antelope than asking
him to wait until you’re done setting up your shot.
50 3D TV and 3D Cinema
The other specialized application for such cameras will be news and talk show
studios with close backgrounds. Virtual studios will love them too, because the
green screen distance is irrelevant, and the FX guys will love to work with a
locked-down camera geometry.
PANASONIC AG-3DA1
When 3D became an established hit in CGI animation movies, 3D aficionados
were on the lookout for any sign that live action production tools were being pre-
pared in research labs. We knew that such research would act as a self-fulfilling
prophecy: if Sony showed a 3D camera, 3D would be a sure business, other play-
ers would have to follow, and Sony would have to make a 3D camera. Indeed,
that 3D camera was actually presented by Panasonic at NAB 2009, in an obscure
corner of the booth, behind the 3D theater. It was presented as a dual-lens
nonfunctional mockup, and was rumored to be a single-imaging system produc-
ing alternated stereo. Response from the audience was twofold. “A 3D camera?
Great, save one for me.” And: “60 Hz alternative 3D? Forget it.” How accurate
was the single-imaging system rumor? We don’t know. And a few months later, a
slightly different product was officially introduced as the AG-3DA1.
There’s an ongoing trend in production equipment: availability of all-in-one cameras has been a key factor
Prices are brought down by higher integration and better for HD deployment in productions. We knew this would
chip designs, and more and more components are put repeat itself in 3D. You cannot put the genie back in
together. In 3D, this will lead to a whole range of spe- the bottle: people want to produce with lightweight cam-
cialized 3D cameras for as many 3D special shots. eras, and that hole needed to be filled in 3D. We knew
it would be easier for us to move up from that point, to
On the production side, the need is to test and try. We
offer a greater range of 3D cameras based on a much
need to invent new compelling shots, and each one may
greater body of experience. Our camera lineup would
require a new camera type. We need that signature shot
grow up with our users’ 3D business and expertise.
that will make 3D worthwhile. To find it, just watch 3D
and be on the lookout. There are a lot of opportunities We sold our first production batch on pre-orders, and will
for clever individuals and companies, because the all- likely do the same on the second batch. This shows the
purpose 3D rig is very far away. The next wave is made same interest from our consumers as for a 2D camera
of specialized rigs. in the same price range.
Based on this, Panasonic looked at the 3D production This camera was used on an all-star game production,
landscape and found that the biggest hole in the offer- along with PACE rigs.
ings was the lack of a single-body camera. For us, it was
the obvious missing key in the acquisition toolset. There To help identify the streams later in the workflow, the
were a lot of 3D rigs of all qualities, prices, and sizes, left and right images on the AG-3DA1 are identified
but there was nothing in the prosumer range. And the with a flag in the HDSDI channels. We are working on
The Stereoscopic Camera Unit Chapter 2 51
SONY
At NAB 2010, John Honeycutt of Discovery Communications briefly presented
a Sony 3D camcorder prototype they were working on. Based on Sony’s best-
selling PMW-EX3, its technology was drawn from Sony Broadcast laboratories
(San Jose), Sony Pictures (Culver City), and Sony Atsugi (Japan), with major FIGURE 2.14
contributions by 3D consultants at 21st Century 3D. It included two camera The Sony 3D Camera
heads, two image-processing pipelines, two digital outs, and a live picture Prototype. Sony intro
beamed on a Sony 4 K 3D SXRD projector. The features list of the camera, as duced this prototype
at NAB 2010, and the
shared by Sony’s Peter Ludé, was as follows:
audience dubbed it “the
n Three CMOS 1/2-inch sensors per eye EX-3D,” based on its
visual similarity with
n Full 1920x1080 progressive capture
Sony’s bestselling 2D
n Interchangeable lens capability camcorder, the PMW-EX3.
n Adjustable interaxial distance (1.5 to 3.5 inches)
n Adjustable convergence via horizontal lens
shift
n Full metadata support
In September 2010, George Joblove, executive
vice president of advanced technology for Sony
Pictures Technology, presented Sony’s vision of
the 3D production system that would surround
this camera. The core concept is to produce a
cheap, lightweight, power-sipping camera, and
to be able to process its footage on a personal
52 3D TV and 3D Cinema
with a single camera. The camera resolution will be cut in half, and one can
chose between time or space resolution.
Time multiplexing is the process of shooting the left and right views alternately
on the full surface of the sensor. There used to be a consumer device called
the NuView designed to record 25/30 fps alternative 3D on 50/60 Hz inter-
laced video systems like PAL or NTSC. Some DIYers have successfully adapted
it on HD cameras, but the optical resolution of its cheap lenses made this a
dead end for HD-3D. A new system, based on an in-lens aperture shift, is cur-
rently being developed by ISee3D. We are yet to see actual pictures generated
by them.
A serious single-camera 3D shoot is done with oversized sensors that still
deliver HD resolution, even if you slice the original frame. There are not that
many cameras that offer super-HD resolution to start with, as the Phantom
and Red do. Fortunately, they boast oversized sensors similar in size to 35
and 65 mm film back. This blends perfectly with historical film 3D that many
3D lens adaptors developed in the 1970s and ‘80s.
FIGURE 2.16
The ISee3D patent for a single-lens, single-sensor 3D camera.
54 3D TV and 3D Cinema
FIGURE 2.17
The Single Camera Mirror Rig Built by Perry Hoberman. FIGURE 2.18
Image courtesy of Perry Hoberman. Vision Research Phantom Camera with a 65 mm 3D Lens.
The Stereoscopic Camera Unit Chapter 2 55
DSLRs IN 3D
What is trendier than shooting art video with
DSLR? Shooting 3D. If only there were a way to do
both. Unfortunately, current DSLRs have no gen-
lock port. This may be added for 3D, but the cost
increase will have to be justified for the majority of
2D users.
The other option is to fit your DSLR with a 3D lens. So
far, the only product available is a Loreo lens mount
with two issues. The first is the resultant image 8/9th
aspect ratio, and the second is the low quality of the
optics that show quite a lot of color fringing.
Recently, Panasonic introduced a 3D conversion lens
for its Lumix G cameras, the H-FT012.
FIGURE 2.20
SINGLE-LENS PARALLAX GENERATION THROUGH APERTURE SHIFTING The Panasonic H-FT012
for Lumix G Cameras.
The single-lens 3D system is among the magic wands that stereographers are Image courtesy of
looking for. Fortunately, there’s a theoretical path to that 3D treasure: the Panasonic.
56 3D TV and 3D Cinema
aperture shift. If you move the iris in a lens laterally, you move the optical axis
around the focal point. If you alternate two equally spaced iris positions in the
horizontal axis, you generate two stereoscopic viewpoints with a single camera.
The issue is that you have to include both apertures inside the lens’ original
maximum aperture. When you increase the stereoscopic base, you do the
following:
n Minimize the left and right apertures.
n Minimize the sensitivity.
n Increase the depth of field.
n Maximize the optical aberrations, working with the edges of the lenses.
Furthermore, you are recording alternative stereo, with left and right images
shot sequentially. Still, this is an interesting approach, and we had a long inter-
view with ISee3D CEO Dwight Romanica about the topic.
We basically inserted a pair of active glasses between In the 1990s there was a 3D lens attachment (NuView)
the lens and the sensor. This allows the blocking of the that can be described as a periscope fitted with a see-
light getting through the left or right half of the lens, vir- through mirror and a pair of 3D glasses. The glasses
tually turning the front lens width into interaxial. In our ran at 60 Hz, locked to the camera’s video out. As a
actual implementations, we are reaching 55% to 90% of result, the left and right frames were recorded on odd or
the front lens turned into interaxial. even fields of standard interlaced video. Seen on a CRT
TV with 60 Hz LCS glasses, a VHS of Hi8 tape would be
On high-end products like broadcast cameras, we will
3D. It was flickering like hell, but it was still 3D. All the
implement a mechanical shutter placed in an optical
current frame-compatible formats are the brain children
attachment. On low-end products like consumer camcord-
of this pre-digital 3D system, whose invention is cred-
ers, we plan to implement a liquid crystal shutter (LCS)
ited to Lenny Lipton. What was possible when all dis-
device that costs pennies in mass production, leading to
plays were running the same frame rate (50i or 60i) is
generalized, single-lens, 3D-switchable cameras.
not possible now, until a strict frame rate and 3D format
The known advantages are single-camera simplicity, per- conversion policy ensure that the stereo is not messed
fect for run-and-gun production, plus the views alignment is up by the time it reaches your eyeballs. Still, the current
persistent, and can even be perfect. Issues are inherent to state of the art is to shoot synchronized left and right,
view-alternated 3D capture, when left and right images are even if the majority of displays in theaters or at home
taken one after another. In a 3D rig, great care must be have a 144th or 120th of a second left/right delay.
taken to ensure that both images are acquired at exactly
the same time. In alternated 3D, you take for granted that At some point, frame interpolation algorithms can be
the images are not taken in synchronicity. If you do a ver- specialized to fix this, knowing they have to interpolate
tical pan, you have vertical disparities. The solution is to movement up to the point at which all disparities are
reduce that delay, or to match the display’s delay. horizontal.
The Stereoscopic Camera Unit Chapter 2 57
The average light loss is announced at one f-stop because only half the expo-
sure time is available per eye. More loss is to be expected in an actual shoot,
due to optical system imperfection, transition time between left and right shut-
ter positions, and a potentially smaller aperture.
A simpler approach is to properly flag the alternatively produced 3D and make
sure the display actually processes the images and presents them accordingly.
On 120 Hz active displays, it’s just a question of appropriately up-converting
the frame rate. Just as we have interlaced and progressive 2D formats that
are taken care of, we would have alternative and synchronous 3D. If such
single-lens active stereo gains ground in professional production, we’ll then see
SMPTE standards describing it. In the analogic world, SD3D (PAL or NTSC)
was supposed to be shot alternatively, and film had to be shot synchronously.
In the digital era, we may see 3D in 1080ia, 1080pa, 1080is, and 1080ps as
combinations of interlaced or progressive frame structure with alternative or
synchronous stereo timing.
FIGURE 2.21
JVC ZCam. In 2003 JVC
introduced this hybrid
camera system with a
Z sensor on the optical
axis.
58 3D TV and 3D Cinema
closest object will illuminate the sensor sooner than ones bouncing on sur-
faces that are farther away. Consecutive frames describe various depth planes.
Illuminating the scene with low intensity radar makes the process compatible
with image acquisition in the visible light range.
Advanced Scientific Concepts, Inc.’s LIDAR system is used for docking the space
shuttle to the ISS, and they are now working on bringing their technology to
the entertainment industry. This technology has a limited resolution, virtually
infinite range, and requires relatively low computing power. It is well suited to
acquire urban landscape shapes, like the geometry of a stadium for 3D conver-
sion systems.
The constructed light approach uses a test-pattern-like image flashed on the
scene. A single image is shot in between two “regular” frame acquisitions and
analyzed. From the deformations of the pattern, the depth of the scene is recre-
ated. This technology has a limited range and requires some computing time.
It is well suited for close-up still image acquisition, and to
generate 3D scans of objects such as actors’ faces. Many
handheld 3D scanners use structured light. Minolta
Advanced
once produced the Minolta 3D 1500, a consumer
Scientific Concepts,
digital still camera that shot 3D images.
Inc.:
www.asc3d.com The SPI depth acquisition records the light
polarization of each pixel. This describes the
Research paper on constructed
orientation of surfaces towards the sensor
light: http://citeseerx.ist.psu.edu/
and the light source. From that point, shapes
viewdoc/download?doi10.1.1.126
are reconstructed, based on the assumption
.6857&reprep1&typepdf
that all surfaces are continuous. Of all these
PhotonX: http://www three depth acquisition techniques, SPI is the
.photon-x.com/ one showing the higher resolution. Photon-X has
developed a set of sensors and plans to integrate
them with broadcast imagers.
CHAPTER 3
3D Image Processing and
Monitoring
59
3D TV and 3D Cinema.
© 2012 Elsevier Inc. All rights reserved.
60 3D TV and 3D Cinema
dedicated processing hardware, with the ability to correct your live 3D stream
as you shoot it, even before it reaches the OB van grid.
MONITORING 3D
The main rule is that you want to see the 3D. Clearly, you need to monitor
your work in 3D. You wouldn’t monitor a color shoot on a black and white
monitor, nor would you record sound without a headset, just by watching the
levels. For the same reasons you wouldn’t do these things, you shouldn’t shoot
3D without being able to see it.
That being said, this does not mean it has to be in stereoscopic 3D. To start
with, not everybody needs to monitor the 3D on a 3D display, and there’s even
a debate on the importance of running dailies in theater to check 3D in its final
screen size. As we’ll see, the best monitoring for stereoscopic images is some-
times on a 2D display. This requires the images to be adequately formatted into
a structure that will reveal their potential defaults.
BASIC 3D MONITORING
Basic stereoscopic control can be done with a 50% mix on a regular 2D moni-
tor. Vertical disparities should be unnoticeable on a small field monitor; other-
wise they’ll be uncomfortable on full-size screens.
Side-by-side and over/under modes can also be used, but they are less efficient
in detecting horizontal or vertical disparities and rotations. Anaglyphic 3D is
often preferred for previews and is best enjoyed without glasses. The red and
blue fringes act as disparity enhancers on the display.
INTERMEDIATE 3D MONITORING
Intermediate level quality control is done using a difference view that
clearly shows the disparities and makes them much more noticeable, even
measurable.
Basic and average controls can be supplemented with markers showing refer-
ence parallaxes on the monitor. One can use simple Post-its or tape applied on
the screen. Some professional displays can electronically insert visual helpers,
3D Image Processing and Monitoring Chapter 3 61
FIGURE 3.1
50% Mix Monitoring. Example of basic 3D monitoring; 50% mix of left and right eye.
FIGURE 3.2
Difference Preview. Example of intermediate 3D monitoring; difference between left and right eye.
like gridlines, bars, or checkerboards. If you are shooting with no target screen
size, 1% and 2% of the screen width are the regular landmarks. If you are
shooting for a defined screen size, the size of the reference grid unit can be
the equivalent of one or two human inter-optical distances. On a feature, the
62 3D TV and 3D Cinema
ADVANCED 3D MONITORING
Advanced controls for specific image default detection—like rotations, key-
stones, and various geometry mismatches—can be performed with the
naked eye and years of experience, or with specialized tools. Alignment
patterns are one option. Image analysis is another. Combining the two is a
great approach.
More complex image processing can be performed on a stereoscopic feed, with-
out going all the way to the complexity of a full 3D analysis. Digital disk record-
ers built around a computer offer some advanced monitoring tools, such as
n Edge detection, for focus matching
n Interactive digital zoom, for fine alignment and convergence
n Alignment grid overlays, for parallax control
n False color zebras
n Spot meters
n Dual histograms, for color balance
n Parallax shifts, to adequately monitor parallel footage
n Anaglyph mixing, to be used with and without glasses
n Wiggle displays, to quickly switch between views
A A
LEFT RIGHT
A VS. A A VS. A A VS. A
LEFT RIGHT LEFT RIGHT LEFT RIGHT
9. Hyperdivergence 10. Edge Mismatch 11. Partial Reverse Stereo 12. Depth Mismatch
A A
LEFT RIGHT
LEFT RIGHT
FIGURE 3.3
The Technicolor Stereoscopic Certifi3D Chart. Another visual representation of the most common
stereoscopic defaults.
Courtesy of Technicolor.
64 3D TV and 3D Cinema
INTER-OPTICAL DISTANCE
Cause: The distance between the cameras is not adequate.
Consequence: In cases of excessive IO, the 3D effect is too strong: the close-up
elements are too close AND the far away elements are too far away. In cases of
insufficient IO, the 3D effect is too weak: the difference between close-up and
far away elements is too small, and the overall scene looks shallow. If 3D is too
strong, it will prevent the audience from enjoying it, generating excessive posi-
tive and negative parallaxes, painful divergence, and toe-in.
Catch it: This is the reason why on-set 3D monitoring is a must. To make it safer
and easier, compute your absolute maximum parallax in screen percentage, based
on your final screen size. Then compute your monitoring screen relative maximum
parallax, in inches or centimeters, based on its resolution and size. Eventually,
draw a rule on a strip of adhesive tape and overlay it on the picture. Check it, or
have it checked as often as needed. If you are using a computer-based recording or
monitoring system, you can get the same results using digital overlays.
Correct it on set: On set, bring the camera to the right interaxial. If you have
reached the limits of your rig, change the camera placement, focal length, or
consider changing the rig. Get a beam-splitter rig for smaller inter-optical dis-
tances, or get a wider base on a parallel rig.
Cure it in post: Inter-optical errors are the most expensive to correct, for one
of the eyes has to be regenerated in CG using 2D/3D conversions, or view-
synthesis techniques.
CONVERGENCE
Cause: Cameras were converged on the wrong point, or one eye was exces-
sively horizontally shifted.
FIGURE 3.4
Simulated IOD Default.
Image courtesy of Steve
Schklair, 3ality Digital.
3D Image Processing and Monitoring Chapter 3 65
Consequence: The whole scene is either too far away or too close to the viewer.
Catch it: As usual, you can catch this by visually checking your 3D, and, if
needed, relying on the visual helpers you draw on the monitoring screen.
Correct it on set: Turn the convergence knob on the rig. If you have reached
its limits, you are shooting under a very strange configuration. Please send me
a picture of the set. If you are sure you need more convergence, shoot with a
huge over-scan, and you’ll reconverge in post.
Cure it in post: This is the easiest, cheapest, and most common 3D correc-
tion. It’s actually a mundane task in 3D to reconverge footage, from on-set to
final grading and packaging. A 3D TV remote should even include a recon-
vergence button to let viewers fine-tune the image to their preferences, which
is done by shifting one or two eyes sideways. However, this will most likely
bring the image edge inside the visible frame, creating an undesired float-
ing window that virtually moves the perceived screen in or out the theater
space. In order to avoid this, a slight zoom in and crop should be applied to
both eyes.
FIGURE 3.5
Simulated Convergence Default.
Image courtesy of Steve Schklair, 3ality Digital.
VERTICAL ALIGNMENT
Cause: On set, one camera is most likely pointing up or down, or a zoom is
off-axis. In post, one eye has been vertically shifted.
Consequence: One of the views is uniformly higher or lower than the other.
This affects the readability of the 3D, not its strength or depth placement.
66 3D TV and 3D Cinema
Catch it: During monitoring, apply horizontal stripes of tape or show guides
in your digital recorder or compositing GUI. Check that left and right images
of objects, shown in anaglyph, 50% mix or difference, are horizontally
aligned.
Correct it on set: If the vertical parallax is more than half a percent of the pic-
ture on a high-end rig, the camera’s geometry should be fixed. On low-end and
makeshift rigs, it may be useless to spend too much time fixing it. This glitch is
easy to fix in post, at the cost of a slight zoom and crop of both eyes.
Cure it in post: Another easy fix: shift the images, just as in a convergence cor-
rection, but this time along the vertical axis. Zoom and crop are to be expected.
FIGURE 3.6
Simulated Vertical Alignment Default.
Image courtesy of Steve Schklair, 3ality Digital.
KEYSTONE
Cause: Keystone appears when the optical axes are not parallel, due to conver-
gence or, less often, strong vertical misalignment.
Consequence: As seen in Figure 3.7, which presents the effects of toe-in con-
vergence, keystone generates asymmetric zooming in the sides of the pictures,
generating vertical misalignment in the four corners. According to stereographic
director Brian Gardner, minor keystone is acceptable and may even help in per-
ceiving 3D, to the extent that our visual system, being itself a toed-in converged
system, experiences and uses keystone as a natural binocular vision byproduct.
This unconventional take on 3D obviously applies to infinitesimal deforma-
tions, not to the huge and painful image distortions that have to be fixed.
3D Image Processing and Monitoring Chapter 3 67
Catch it: Using the 50% mix and the horizontal lines, you’ll see a picture that
is okay in the center, but is symmetrically skewed in the corners. The top and
bottom of the center third of the picture are okay.
Correct it on set: Maybe you should not converge on that shot. Shoot parallel,
get a huge over-scan, and set the convergence in post.
Cure it in post: Fixing a keystone in postproduction is not an easy task, but a
small correction is easy to set up and will be efficient. Apply a perspective dis-
tortion in the direction of the offending keystone, and then zoom and crop as
usual. A significant amount of keystone requires complex depth warping and
falls into the expensive fixes one should avoid. For shots that will be mixed
with CG images, stereoscopic supervisors tend to prefer to work with footage
shot parallel and then reconverge in postproduction.
FIGURE 3.7
Simulated Keystone Default.
Image by Bernard Mendiburu, based on Steve Schklair, 3ality Digital.
ROTATION
Cause: Image rotation appears when the camera’s optical axis is rotated along
the Z axis.
Consequence: Horizontal alignment is progressively lost from the center to
the four corners of the image. It generates painful vertical parallaxes on the
sides and depth artifact in the center top and bottom areas.
Catch it: On the monitoring screen, the image is okay in the center and gets
worse as you approach the edges. All four edges of the frame are off, in asym-
metrical directions.
68 3D TV and 3D Cinema
Correct it on set: There’s something wrong with the rig, most likely a rigidity
issue on a makeshift rig. If it’s significant, it may be the sign of a mechanical
default in the rig mechanical integrity, and therefore should be addressed as a
safety warning. Sometimes, tension on a cable or momentum on an attachment
changes the balance of one camera.
Cure it in post: Image rotation can efficiently be fixed in post, at the cost of a
crop and resizing.
FIGURE 3.8
Simulated Rotation Default.
Image courtesy of Steve Schklair, 3ality Digital.
FOCAL LENGTH
Cause: Prime lenses are not always matched and zoom lenses are almost
never matched, and therefore it is difficult to get two cameras zooming sym-
metrically. If you zoomed with a low-end rig using prosumer cameras, you will
likely have to fix it in post.
Consequence: One image is bigger than another, the center is okay, and the
3D gets worse at the edges, with asymmetrical depth artifacts in the mid-high
left and right areas.
Catch it: The 50% mix shows a picture matching in the center but not on the
edges. The misalignments are radially symmetrical, from the image center into
the four directions.
3D Image Processing and Monitoring Chapter 3 69
Correct it on set: Correct the focal length on one camera. If you are using a
low-end camera with motor-only control of the zoom, you’d better save the
trouble and plan to fix it in post.
Cure it in post: This is another glitch that’s easy to cure in post. Because most
other corrections include a resize of the images, focal length adjustment usu-
ally comes for free in the touch-up package.
FIGURE 3.9
Simulated Focal Length Default.
Image courtesy of Steve Schklair, 3ality Digital.
FOCUS MISMATCH
Cause: Low-end cameras have poor manual control on focus distance, and
autofocus modes may disagree on the subject distance. Despite the efficiency of
our visual system at pasting sharpness from one eye to the other, we eventually
notice it when focus asymmetry gets too big. In CG imagery, it sometimes hap-
pens that one eye has no motion blur pass. This causes a strange feeling and
adversely affects the fusion of the images. Focus matching is a visual quality
element to be taken care of in 3D. The relation between the focus distance and
focal length with zooms is another point of concern.
Consequence: One image is sharp, the other is blurry. Only certain depth
ranges of the pictures are affected.
Catch it: Focus asymmetry is really hard to catch on a 50% mix and is better
seen on a side-by-side view. On a 3D display, you will have to alternately close
each eye and compare the images.
70 3D TV and 3D Cinema
Correct it on set: High-end rigs with professional lenses and remote focus
should never give you this trouble. On low-end rigs, this is much more com-
mon, and there’s no real cure if your cameras have no focus ring. In this case,
you have one more reason to get more light and an infinite focus rather than a
shallow focus.
Cure it in post: Obviously, this has to be taken care of on set. In postproduc-
tion it is close to impossible to regenerate sharpness; the sharper eye would
have to be blurred to match the other one. A 2D/3D conversion may be the
only solution.
FIGURE 3.10
Simulated Focus Mismatch Default. Looking at known sharp textures like human faces helps detect
focus mismatch.
Image courtesy of Steve Schklair, 3ality Digital.
DESYNCHRONIZED STEREO
Cause: The two cameras are not running in perfect synchronism. If both left
and right images are shot without paying great attention to time synchroni-
zation, issues will definitely occur. The faster the camera or the action moves,
the higher the timing requirements. If your image is motion blurred, you need
an extremely accurate genlock, down to the line or pixel clock, or the motion
blurs will have vertical parallaxes.
Consequence: The slightest time delay will transpose camera and actor move-
ments into unwanted parallaxes. If you took care of every mismatch listed ear-
lier in this section, but missed the time synchronization, you will get shots
with all the defects you tried so hard to control.
3D Image Processing and Monitoring Chapter 3 71
Catch it: Objects that have a cyclic movement, like a waving hand, are the best
cues for incorrect timing. They show inconsistent parallax; the action in one
picture is constantly catching up with the other.
Correct it on set: Check this with the director of photography, the camera
technician, or the first, second, or third assistant—whoever is in charge of the
genlock. If you are the person in charge, check the clock distribution and check
that everything is perfectly symmetrical, down to the quality and the length of
the cables.
There are some cameras that can be genlocked at video frame rates, like 25 or
30 fps, but not at the digital cinema 24 fps. The Sony PMW-EX3 has this reputation.
Cure it in post: Try retiming software, or some time warp filters from your
favorite FX suite. While you are computing optical flows, you may give a
chance to 2D/3D conversion too.
FIGURE 3.11
Simulated Synchroni
zation Default.
Fast movement can
generate vertical
mismatch, as on the
waving hand, plus fake
depth, as on the jumping
dog.
Image by Bernard
Mendiburu, based on a set
from Steve Schklair, 3ality
Digital.
have your scale model windmill with you? The director is likely searching for
you with a baseball bat. Try to secure it and have someone swing in front of
the camera. That should do the trick.
Correct it on set: Turn one of the cameras upside down.
Cure it in post: Same as synchronization mismatch, besides the fact that you
need to apply a vertical gradient to the retiming. However, we are not aware of
such a feature in correction tools.
Beware: Active display of passive stereo and vice versa can interfere with this
issue.
3D Monitoring Tools
We are turning a corner, moving from makeshift
3D monitoring into professional gear. Not long ago, monitoring 3D required FIGURE 3.12C
processing left and right images with a dedicated computer and plugging it Simulated Iris
Mismatch Error.
in on one of the very few 3D displays available. Rear projection DLP TV came
Detecting lightness
first, and passive polarized monitors followed more recently. The situation is mismatch requires a
now totally different with the following new products: side-by-side display
and a very sharp eye.
n Consumer 3D displays with no dual input and image processing. Image courtesy of Steve
n Professional 3D displays with integrated dual input and some 3D processing. Schklair, 3ality Digital.
n Simple 3D multiplexers that get two images into a 3D format.
n Complex 3D processors with some image-manipulation capabilities.
CONSUMER 3D DISPLAYS
Consumer 3D displays have the ability to show 3D, but they were not designed
to be used on set or on location. They cannot run on batteries, and most likely
require the 3D images to be processed for display. They all cost from $1000 to
$5000, with the plasmas, full HD projectors, and passive TVs in the upper margin.
If you are using these displays, remember to thoroughly investigate the setup
menus to disable any image enhancement features like smooth motion and
other cinema colors. The frame interpolation, which adapts the 24 fps input
into the display’s 120 Hz refresh rate, is a known culprit. It is usually built
around a modernized 3:2 pull-down to generate a pair of 60p video streams.
PROFESSIONALS 3D DISPLAYS
Professional 3D displays were specially designed for use on set and on loca-
tion. Most of them run on batteries and have left and right input and SDI and
HDMI connectors. Some have image-processing features for monitoring 3D,
such as 50% mix, left/right difference, and reconvergence. Some monitors offer
flip and flop image inversion to monitor 3D coming live from a mirror rig.
Another useful function is an overlaid grid with variable pitch that allows for a
quick check of excessive disparities.
Transvideo was the first company to offer a field monitor. The CineMonitorHD12
3DView is a classic 12-inch with left and right SDI inputs, and anaglyphic and
active stereo modes.
The second generation of professional 3D field monitors use passive 3D dis-
plays, with polarizing filters that rotate each line in opposite directions. One
can look at the 3D picture with low-end disposable 3D cinema glasses, or with
high-end designer glasses that offer a much better contrast.
Panasonic has a 25.5-inch 3D monitor, the BT-3DL2550, with dual SDI and
DVI-D inputs and a 1920x1200 resolution. It displays 720p, 1080i, and 1080p
video in 10 bits color depth.
Panasonic plasmas are to be considered too. They offer full res-
olution, thanks to active 3D, high brightness, saturation, and
contrast. The stereo separation is great, with no ghosting or
sweet spot effect. Professional models for field use are not yet in
catalogs.
Sony offers the LMD2451TD, a 24-inch monitor with the same
1920 3 1200 resolution. The LMD4251TD is a 42-inch version.
JVC has a 46-inch monitor, the GD-463D10U, and at NAB 2009
presented a 4 K 3D monitor prototype.
Assembling two monitors in a beam-splitter mount will give you
a full-resolution passive display. Cinetronic, among many others,
offers such a system. It was used in the production of Pirates of the
FIGURE 3.13
Panasonic 3D Monitor. Caribbean 4.
analyzers. They are working at interfacing their computers with other ven-
dors’ motion controls.
COMPUTER-AIDED MONITORING
Stereoscopic image analysis makes it easier to live with the truth that one has
to see 3D to make 3D. The problem is, of course, that you may not always,
if ever, have a good enough 3D display on set. Let’s talk size: even the best
professional 3D displays will not give you the finesse you need to evaluate
an image that’ll be seen on a 40-foot screen. Current 3D monitors come in
10-inch, 22-inch, and 46-inch sizes, on which evaluating parallax, divergence,
and vertical disparities is quite a challenge. Experienced stereographer Vince
Pace explains that a 46-inch 3D screen is all he needs to evaluate 3D imagery.
Before you get to that level of experience and confidence, however, you will
benefit from the help of dedicated 3D vision computers.
Another problem with field 3D monitoring is brightness. 3D visualization sys-
tems have an overall efficiency ranging in the 15% to 20% range, which means
that with your 3D glasses on, you’ll get a picture five to six times dimmer than
you’ll get on an equivalent 2D display. Here are the visual modes you’ll find on
a typical 3D-Computer Aided Monitoring:
1. Show accurate parallax, vectors.
2. Identify screen plane, foreground, and background using color-coded tags.
3. Show the depth bracket of the shot on depth histograms.
4. Compare the measured parallax against the target screen size.
5. Run various 2D image quality controls in a 3D mode.
Disparity Vectors
In this mode, homologous points are shown with an arrow linking them.
This mode allows for evaluating the parallax and pinpointing most geometric
defaults, such as rotation and the causes of various vertical disparities, including
zoom mismatch.
78 3D TV and 3D Cinema
FIGURE 3.15
Disparity Vector
Monitoring. Disparity
vector monitoring on
disparity tagger.
The Slava Snow Show;
image courtesy of Binocle
and La Compagnie des
Indes.
Disparity Tags
In this mode, homologous points are shown with false color points or icons
showing the amount and direction of parallax. A cross-vendor color-code
convention settled on blue tags for out-of-screen elements, green tags for the
screen plane, and red tags for the infinity plane. Elements placed in the spaces
between these arbitrary planes are in shades of cyan and yellow. Purple is
sometimes used for flagging objects with a negative parallax stronger than the
desired maximum foreground.
FIGURE 3.16
Disparity Tag Moni
toring, with Depth
Bracket. In addition to
discrete disparity tagging,
this monitoring mode
displays, on the lower
third, a global depth
placement of the scene.
The Slava Snow Show;
image courtesy of Binocle
and La Compagnie des
Indes.
3D Image Processing and Monitoring Chapter 3 79
The threshold is computed upon dialed-in screen size, and sometimes addi-
tional factors including viewing distance and divergence limit. If you prefer,
you can type in your reference positive and negative parallax values in pixels or
percentage of image width.
Binocle’s Disparity Tagger displays geometric shapes that enhance visual
information. Strong positive or negative parallaxes are shown with trian-
gles pointing upwards or downwards. Screen plane is tagged with squares.
Vertical misalignment is shown by rotating the shapes to align their base
with the corresponding vector they represent. The size of the shapes repre-
sents the confidence of the computation. The areas of best efficiency happen
to be the subject of interest, with sharp and detailed textures. In other words,
your hero is more likely to be strongly tagged than the background—how
convenient!
FIGURE 3.17
Depth Analysis on 3ality SIP. In this monitoring mode, many parameters are tracked, and a visual representation of the “depth
budget” is shown in the lower third.
Image courtesy of 3ality Digital.
FIGURE 3.18
Corrected Frame
Monitoring on Binocle
Disparity Tagger.
Comparative 2D Analysis
All the image analysis tools you used in 2D will be needed in 3D too, not only
to analyze the 2D images for defaults, but sometimes to check for 3D coher-
ency. Typically, sharpness or edge detection is useful to check the focus and iris
match between cameras. Overlaid vector graph and RGB histograms are useful
tools for matching the light level and color balance on cameras.
FIGURE 3.19
Dual-Channel Color
Histogram and 3D
Vectorscope on 3ality
SIP.
3D Image Processing and Monitoring Chapter 3 81
LENS PAIRING
Lenses, mirrors, filters—there are many elements on the light path that are not
digital, and therefore may never be perfectly matched. Quantifying the discrep-
ancies on set will give you a chance to correct for as much as possible before
the raw picture gets recorded and the stereo balancing has to be done in post.
Do a favor to the FX team: get the most you can from the lens pairing. After
that point, all corrections are time, money, computers, and long nights.
82 3D TV and 3D Cinema
Z-LUT Generation
Once the rig is aligned, it’s time to take care of the lenses. Some lens vendors
offer matched zooms sold by pairs. At some point, lenses may even have their
LUTs computed at the factory and embedded somewhere in an on-board mem-
ory. In the meantime, you’ll be working with lenses that do not match, and you’ll
have to take care of that, in shoot preparation and in postproduction. Just like
color matching, you’d have a much better time in post if you did it well in prep.
Before image analyzers stepped in, Z-LUTs were generated by zooming all the
way in and out on an alignment chart. Because zoom telecentry follows vari-
ous sinusoidal curves, you need to realign the cameras at as many in-between
points as needed, until the motion-control system tracks it closely enough.
Automatic Z-LUT is a great tool that shrinks this process from 30 minutes to a
handful of seconds.
Ongoing improvements in technology will soon allow you to get rid of align-
ments patterns. This is important for live productions, when the zooms are
known to drift away and their LUTs may have to be recalibrated. Thermal dil-
atations are the most obvious issues, for example, following nightfall during
soccer matches or heat generation by dancing crowds in a concert location.
FUTURE APPLICATIONS
At the moment, the industry is only scratching the surface of the potential
benefits of image analysis. With the increase in computing power and finessing
of the software, more tools will soon come to fruition.
Roundness Evaluation
On a 2D TV, all balls are round. In a 3D TV program, some are spherical, flat,
or elongated, all depending upon the screen size, viewing distance, depth posi-
tion, and inter-optical distance. Because of the effect of the viewing distance,
it is not easy to set the inter-optical and convergence solely based on a visual
check on a small screen. At some point, image analysis will be able to commu-
nicate to you an evaluation of the “roundness” of a shot by identifying known
objects like human faces, and comparing their 2D on-screen size with their
depth range.
Existing Products
There are currently four high-end products on the market, from 3ality, Binocle,
Sony, and HHI/KUK. (We will discuss this more in the next section, on image
correction.)
3D Image Processing and Monitoring Chapter 3 85
FIGURE 3.20
Cel-Scope3D Analyzer.
okay, until the correction system kicks in and you feel like someone turned the
light on while you were reading in dim light.
It may seem like overkill to correct what appears as nothing more than visual
background noise—but, in reality, there’s no such a thing as visual noise in
3D. What you get are retinal disparities, increasing the audience’s visual system
workload. Making sense of this imperfect 3D requires physical realignment at
the eye level, and additional fusion complexity in the visual cortex. This means
causing visual fatigue for the fittest, and headaches for the masses—not a good
selling point for your 3D content. You may think that a couple pixels of offset
are not enough to generate real discomfort, but you’d be incorrect.
In addition, as soon as you mix graphics with your images, you mix perfect
geometry—the CGI—with imperfect geometry—the live feed. In a post-
produced feature, it means hours of labor to fine-tune it, to make the blue-
screened actors match the CG backgrounds and stick on the floor. In a live 3D
feed, like a sporting event, it means that the audience has to choose between
looking at the scores or at the players.
Post tools will be presented in Chapter 4, and live correction tools are listed
hereafter.
Scene depth compression would deal with excessive IOD by bringing back-
grounds and foregrounds toward the screen plane. This is possible because
there’s no occlusion revelation. Still, the edge detection needs to be perfected.
Object depth extension would deal with insufficient IOD by inflating the volume
of objects. This would require increasing the accuracy and resolution of depth
maps generated from the disparity vectors.
Existing Gear
3ALITY
The first to market was 3ality Digital, with its SIP 2100 and the upper-scale,
multicamera unit, the SIP 2900. It offers extensive monitoring modes, DVI
and SDI connectivity, frame-compatible format conversions, and Web remote
3D Image Processing and Monitoring Chapter 3 89
control. It also drives the rig motors and corrects geometries. It does all this
using two PowerPCs and four Xilinx Spartans, for a total processing power of
a teraflop. It uses mostly 2D image analysis to control the cameras and make
sure the best image is produced from the start, while 3D geometries are com-
puted and corrected in non-Euclidian spaces.
This piece of equipment has been under development for many years and
is used on 3D live productions like BSkyB Sports, and on movie sets. Even
though it was designed to work with 3ality rigs, it is often used with other sys-
tems, especially on feature productions.
The 3ality approach is to put the quality burden on the rigs and cameras, and
to minimize the amount of corrections to be applied afterwards. These qual-
ity expectations are to correct “mechanical shifts” that will always occur during
the production of a live show. When you keep zooming in and out, and pan
the camera side to side for hours, you apply momentum on cameras, rigs, and
lenses. Experience shows that your mechanical registration will not last until
the end of the show. To that end, the image has to be analyzed and corrections
have to be sent to the rig.
BINOCLE
Binocle is a French company that has been doing stereoscopic research and
productions for more than ten years. They have teamed with the French
National Institute for Research in Computer Science and Control (INRIA),
known for creating the RealViz FX suite. At NAB 2009, Binocle introduced its
image processing system, the Disparity Tagger. It is a regular Windows com-
puter with SDI in and out capabilities. At NAB 2010, the first generation of
real-time geometric correction was presented. Binocle’s philosophy is to embed
the rig’s motion control and computing resources inside an all-in-one socket
that holds the camera. As a result, the image analyzing system is totally inde-
pendent and does not interact with rig automation.
KUK
The STAN is the result of collaboration between the Fraunhofer Heinrich Hertz
Institute (HHI) and KUK productions. It was presented at NAB 2009 and has
since been adapted for P S Technik rigs. It is now available.
The STAN approach is to use information coming from the rig’s motion con-
trol to help with the time extrapolation part of the geometric correction. It’s
also a software solution that runs on a PC platform. It has a simplistic but
intuitive user interface.
SONY
Sony’s product is composed of a multipurpose digital production system, the
MPE-200, and a 3D acquisition specialized software, the MPES-3D01. The
hardware is based on the Cell BE2 engine, with four HD inputs and outputs,
and the software is developed in the United States.
90 3D TV and 3D Cinema
FIGURE 3.21
Camera Alignment on
Sony MPE-200 with
MPES-3D01.
91
3D TV and 3D Cinema.
© 2012 Elsevier Inc. All rights reserved.
92 3D TV and 3D Cinema
FIGURE 4.1
Camera Unit on Derrière les Murs (Behind the Walls). In this picture the rig is in a “looking up”
configuration, freeing the operator’s line of sight.
Shooting 3D for Broadcast or Editing Chapter 4 93
The last point to consider is the dust. On a down-looking rig, dust can fall on
the mirror and adversely affect the image. On a 3D TV show or documentary, the
depolarization filters will help protect the mirror from environmental hazards.
DSC’s major technological breakthrough over typical test charts is precision and increased dynamic range. In addition to DSC’s
patented grayscale and color elements, this chart provides many unique features specific to 3-D. DSC charts are designed to pro-
duce familiar electronic displays that are easily interpreted using standard waveform monitors and vectorscopes. In production, a
few frames shot on set provide invaluable information in post.
Shirley’s Webs -
Vertical Scales -
enable meticulous DSC Jet Black Background -
on right and left, assist in
camera matching provides clean display
checking rotational offset
This chart can be invaluable in aligning 3-D camera systems - while everyone has their own methods, here are some basics:
Align cameras so that images match exactly for grayscale levels, tracking and color reproduction. Adjust horizontal, vertical
and rotational positioning, image size, focus distance, resolution etc. The aim is to match superimposed images as closely as
possible, then set interocular and convergence to the particular requirements of the scene.
Shooting a few frames of this chart on the set during production can be invaluable in providing points of reference between
cameras for use in Post.
QMF26
Tel. 905.673.3211 - e-mail [email protected] - website www.dsclabs.com January 2011
FIGURE 4.2
The DSC Labs CamAlign 3-D Pilot. This chart was designed in collaboration with Sean Fairburn to help align cameras and correct
geometries in postproduction.
Shooting 3D for Broadcast or Editing Chapter 4 95
Use a 3D monitor in anaglyph viewing mode to perform fast and precise alignment.
The anaglyph viewing mode is a robust tool to monitor all stereoscopic parameters
during shooting.
1. Check camera orientation. In order for the rolling shutters of the two digital cameras
to operate in the same direction, make sure one of the cameras is mounted upside
down. Tango provides dedicated inverter blocks for the Mini 2K SI, the Red One, the
Sony HDC-P1, the Sony F23 and F35, and the ARRI Alexa.
2. Install and lock both cameras with lenses onto the 3D Stereo Tango rig, and set the
interaxial distance to zero. Keep the interaxial at zero during the entire alignment
checklist.
3. Check the distance between the beam-splitter and the two cameras. Set the adjustable
stop blocks at the same value for both the left and right eyes, and make sure each
camera is pushed against its respective stopping block.
4. Check the vertical alignment of the optical centers. Look into the matte box straight
through the beam-splitter with one eye only; the two lenses should overlap, and
there should be no vertical offset. If there is a vertical offset, use the large silver
thumbwheel behind the top camera to move it up or down until the two lenses
overlap perfectly along the vertical axis.
5. Check tilt alignment. Looking at a distant landmark on the 3D monitor, use the tilt
adjustment knob on the bottom camera to make the left and right images overlap
perfectly on the vertical axis. Then check this alignment looking at an object five
feet away from the rig. If there is now a vertical offset between the left and right
images, adjust the optical centers as described in Step 4, and re-do Step 5.
6. Check roll alignment. Look at the horizontal structures on the 3D monitor. If on the
left side of the picture a red border above light-toned elements appears, it means
the bottom camera (left) leans to the left. Using the two silver knobs located on
either side of the roll stage under the bottom camera, adjust the roll until the left
and right images overlap perfectly across the frame on the vertical axis.
7. Check convergence alignment. Look into the matte box straight through the beam-
splitter with one eye only; the two lenses should overlap, and there should be no
horizontal offset. If there is a horizontal offset, use the convergence knob beside
the bottom camera to adjust the angle until the two lenses overlap perfectly along
the horizontal axis. Then look at a distant background object on the 3D monitor,
and turn the convergence knob until you make the left and right images overlap
perfectly on the horizontal axis. The left and right cameras are now parallel.
8. Check viewing angle alignment. If you’re not using perfectly matching prime lenses,
you may notice a slight magnification disparity on the 3D monitor. Point the rig at
a round object: the object will appear slightly larger in one eye than in the other.
This is a magnification disparity, which translates into a disparity of viewing angles
between the two lenses. To suppress this disparity, move the bottom camera closer
or farther away from the beam-splitter until, on the 3D monitor, the magnification is
the same for both eyes.
Note: For fine-tuning purposes, all of the adjustments described above can also be
applied to the top camera. The 3D Stereo Tango provides six independent axes of
alignment for the top and bottom cameras.
The roll, tilt, and convergence adjustments all pivot on the beam-splitter at the intersecting
points of the optical axes of the left and right cameras. The roll, tilt, and convergence
adjustments can be done manually, or they can be motorized.
96 3D TV and 3D Cinema
FIGURE 4.3
Stereo Tango Mirror Rig.
Stereoscopic HD-SDI
“Duplicate everything” is the basic modus operandi of stereoscopic produc-
tions. Cables are no exception to this rule, with the complexity one can imag-
ine. However, in some cases, especially for remote camera positions, you’ll
want to run single cables.
As a reminder, we’ll describe the serial digital interface family of standards at
HD resolutions:
n SMPTE 292M: the legacy HD-SDI, 720p and 1080i resolutions
n SMPTE 372M: the dual-link HD-SDI for 4:4:4 encodings and 1080p
resolutions
n SMPTE 424M: the 3G HD-SDI, basically two SMPTE 372M on a single cable
Although 3G SDI can be used to actually transport stereoscopic streams, it
has not been standardized yet and has been dubbed 3D over 3G. Still higher
bitrates are needed, and they will be standardized by SMPTE for stereoscopic
4:4:4 encoding, RGB color space, 12-bits color depth, and higher frame rates in
progressive modes. In the meantime, we rely on vendor solutions to multiplex
image streams on single links.
Stereoscopic production was an early adopter of fiberglass cabling for signal
transport. At IBC 2010, Telecast Corporation presented a fiber converter that
Shooting 3D for Broadcast or Editing Chapter 4 99
Frame-Compatible Formats
Stereoscopic production reuses a lot of 2D gear and setups. In some cases, 2D
tools have enough bandwidth and connectors to process two HD steams, and
a clever configuration or a software update makes them 3D. Sometimes two of
them are synched and used in parallel; sometimes you must deal with a single
HD channel capacity to transport or process 3D. In these cases, the 3D will be
encoded in a 2D frame-compatible format (FCF).
BASIC FCF
The basic process of frame-compatible formatting is to drop half the resolution
and pack the remaining anamorphic versions of the left and right images into
a single 2D image. Most basic FCFs have the self-explanatory side-by-side (SBS)
and top-and-bottom (TAB) image arrangement. A third format is the line inter-
leaved form, with horizontal lines alternately describing the left or right images.
This is the display format of the current generation of passive 3D TV, which
uses Arizawa pattern polarizers. This was also used in analog TV to record left
and right views on separated fields. For this reason, you’ll see interleaved FCF
3D referred to as “passive” in recent display literature, and “active” in older
material. Both are equally inaccurate.
The main problem with FCF is obviously the loss of resolution, and the value
of each format results from the direction of the resolution decimation. The
side-by-side seems to be preferred for many reasons; for example, experienced
stereographers can watch it in 3D using the “free viewing” technique that con-
sists in voluntarily converging or diverging their eyesight until the left and right
images overlap and the stereoscopic images fuse into a 3D one.
For interlaced 1080i video, SBS is better than TAB, for each frame still has
540 lines rather than 270. For progressive 720p video, TAB is better than SBS,
because the 2D horizontal resolution actually defines the depth resolution.
Various standardization groups explicitly recommend or impose SBS as the
1080i format and TAB for 720p, with the left image on the left or top part of
the frame, respectively. This rough process is especially efficient for recording
and transporting 3D. A frame-compatible formatted 3D can be routed through
thousands of types of 2D equipment and will never get out of sync.
The FCFs are transition formats that are expected to disappear from the profes-
sional universe as the 3D transition progresses. They are likely to have a longer
life in consumer electronics, where their main application is the HDMI 1.3a
standard that embeds 3D signals into a single HD link from the Blu-ray Disc
100 3D TV and 3D Cinema
player to the TV. The 3D conversion lenses generate SBS FCF that is captured
and recorded as it is on the support, and eventually sent to the TV accordingly.
The main concern with FCFs is the danger of seeing them stacked atop one
another. If an SBS and an TAB FCF are used consecutively, the resulting signal
is a quarter HD, marginally better than SD.
ADVANCED FCF
Frame-compatible format quality can be optimized with a smarter decimation
scheme: the quincunx or checkerboard format, also called the mesh format.
A quincunx (Qx) 3D image is composed of pixels alternately taken from left
and right images in a checkerboard pattern. This format is also known as DLP,
because it has been used in 3D-capable RPTVs since 2007. It exists in two
types: native Qx, with left and right images composed together, and rebuilt
Qx, with left and right images reconstructed on the left and right parts of the
frame. A native Qx looks like a 50% mix, and a rebuilt Qx looks like a side-by-
side picture. Both present high frequencies. A native Qx is barely compatible
with any digital compression, whereas a rebuilt Qx will just require a higher bit
stream than a regular side-by-side to preserve its additional visual resolution.
This subsampling allows for a much better image reconstruction, because each
missing pixel can be rebuilt from four neighbors, rather than only two, and
aliasing artifacts are less noticeable.
A proprietary version of the checkerboard was patented by Sensio. Their pro-
cess relies on pre-filtering the images and post-processing them to regener-
ate even more visual resolution. The claim is that 90% of the resolution is
Shooting 3D for Broadcast or Editing Chapter 4 101
preserved. The Sensio encoding is recognized as the most advanced and effi-
cient FCF format by the transport industry, and is deployed in most transport
providers’ facilities offering 3D services.
FRAME PACKING
The two streams of 30fps stereoscopic content can be stacked together into sin-
gle 60fps monoscopic format. That’s what Digital Cinema Initiatives (DCI), the
joint venture of the major Hollywood studios, does with 3D movies encoded
and transported as 48fps content. It is crucial to properly flag the frames in
ancillary metadata, so that the left and right images are properly restituted.
Otherwise they’ll be out of sync by one frame.
DIRECTING LIVE 3D
Most of the experience in live 3D production is in sporting events, which
clearly lead the industry, with concerts following behind. Sports broadcasts fol-
low a ritualized storytelling process, with the main cameras on high ground
doing up to 80% of the air time, as well as close-ups on some actions, and a lot
of replays and graphics. A 3D production tries to revisit the TV sports liturgy,
bringing its own wow shots without breaking the magic and the story.
If the audience says “The 3D was great, but we didn’t get the game,”
we failed.
Vince Pace
Finding the Right Balance Between Showing You Have to Slow Down Directing There are many more
Off and Game Presentation traps in 3D than in 2D. In 2D, I can just see a shot that
Derek Manning directed 3D TV football and baseball is good, even if it’s down in a corner on a black and
games for Fox. Here’s what he shared with us during an white monitor, and I can take it: Go! On a 2D football
interview. game, cutting is quite hectic.
You can’t do that in 3D, or it’s going to be hard on the
Camera Positions Are Key to Good 3D Sports Finding
audience’s eyes. You don’t want to do it for two reasons:
the best camera positions is key in doing good 3D
first, because you should preview that shot. Ask for, say,
sports. If you’re too high, you won’t get a good 3D
“ready seven,” and the TD will preview it for you on the
shot—and if you’re too tight, you’ll get great 3D, but you
3D monitor, allowing you to check it in 3D before you
won’t get the best views for showing the strategy of the
take it.
game. For the high and wide play-by-play shots, you want
the cameras to get lower and closer to the field, in a Second, because every shot is an experience in 3D.
medium-high shot position. There’s so much more in a good 3D image—you don’t
102 3D TV and 3D Cinema
need to cut as much as in 2D. I remember watching one cameraman can see this in advance, and alert the direc-
3D shot during a show, and I was so drawn into it that I tor on the intercom so he can get off that shot.
started to watch it like TV, almost forgetting I was actu-
The Shots You Want in Your Show Nice 3D shots have
ally directing.
many layers of depth, with people or objects filling the
Camera Operators Have a Learning Curve Too Plan for space. This is especially true in baseball fields, where
longer camera meetings, because your crew will need to each player has a separate position and they are spread
learn new things; they are experiencing the learning pro- out, with the crowd behind them. In American football,
cess too, just as we are in the truck. they are more packed together. Still, there are touch-
You want them to show off the 3D; you want the higher, down shots with a player holding the ball at the camera,
wider camera to shoot a little tighter than normal, the other players running toward the cameras, and the
and the lower camera on the ground to shoot wider. stadium in the background. This kind of shot has layers
Normally, these cameras would be right on the guy’s of depth—foreground, middle ground, and background—
face, but you may need to ask for a head-to-toe or waist- making for gorgeous 3D shots.
up shot—because without the background behind the On the other hand, tight shots like those we often do
player, it’s not really possible to see the 3D. Additionally, in sports will not come out great. In 2D, it’s common
it’s hard on stereographers to get their 3D correct when to do a close-up on the quarterback’s eyes. In 3D, you
you are shooting that tightly. won’t get any depth effect from that shot. However, it’s
Teach them to watch out for edge violations, like people a compelling image that you want, no matter what; this
entering the frame, or bright objects on the sides of the is where you need balance between the 3D show and
frame. They have to get rid of that, one way or another. the game. I remember the stereographer would warn
Get it in the frame, or get it out—but deal with it. You me some shots would not make good 3D. He came with
may also need their help with this for some camera a movie perspective, which was his background. I was
positions; for example, on a low-end zone shot in a foot- doing those head shots, and he was shaking his head.
ball game, if a player is coming up the field, he will cross I told him, “I have to do that, that’s capturing where the
the frame, which is going to be hard on your eyes. The game is—on his face.”
3D CONVERSION
Real-time 3D conversion is making great progress in field use. Part of its suc-
cess comes from the scarcity of 3D camera positions, or the inability to pull
good 3D from specific shots. Along with being driven by demand, real-time 3D
conversion benefits from the current audience’s forgiveness for poor 3D.
Following the mantra that “the show must go on,” let’s see three cases where
real-time 3D conversion can be used with acceptable results. The quality of the
conversion will be inversely related to the intensity of the generated depth.
n Large landscape with little or no discernable depth features.
n Medium landscape with known geometry.
n Deterministic 3D conversion following known “world rules.”
away from the camera do not generate any occlusion revelation, and therefore
they appear in the same depth as the ground or seat they are on. A properly
acquired 3D shape of an empty venue that is then fed into a conversion system
will generate good stereoscopy.
The only issue with this technique is that there isn’t currently a vendor offer-
ing this as a turnkey solution. Still, it’s not a complex effect to put together, and
it can be simplified by generating a depth map and using it as a displacement
map source. If you can synchronize the depth map to the camera movements,
either via motion control or via point tracking in the image domain, you’ll get a
dynamic 3D conversion. More advanced processes, including depth-acquisition
devices and CG modeling and projection techniques, will perfect it.
domain, it will generate some magnification and cropping. The most proba-
ble implementation we’ll see will be in the video switcher, using depth meta-
data embedded by image analyzers in the ancillary channel, which will mostly
reconverge the out shot for faster execution.
3D OUTSIDE BROADCAST
Prepping an HDTV OB vehicle for 3D production is technically easy. You have
three options:
n Keep it pristine, walk in, and produce frame-compatible 3D.
n Temporarily retrofit an existing HD van for 3D production.
n Rent one of the brand-new 3D/HD trucks.
Your main problem may be to find the place for the stereo staff: typically the
stereographer in the main truck, and the so-called convergence pullers groom-
ing the 3D feeds in a second truck.
Producing Frame-Compatible 3D
If you have no choice other than using an HD truck with no budget to convert
to 3D, you’ll have to go the FCF route. First, select the FCF you want to use—
most likely SBS, for its better compatibility with interlaced video; or quincunx
for higher reconstructed resolution. Encode all your sources, including graph-
ics and titles, in that FC format. Run your camera feeds through 3D multiplexers
and process them as 2D. If you work in quincunx, make sure there are no issues
or artifacts with video compression, encoding, or processing, as you would do in
instant replay systems.
Select transitions that process the images in a way that is compatible with the
FCF you have selected. The raw quincunx has the good grace to be compat-
ible with most transitions. (Sensio is among the very few companies that has
developed a real knowledge in such “soft retrofitting” processes.) Most 3D dis-
plays will be able to decode the FCF on the fly for your program and preview
monitoring, but some may need an external decoder, especially for quincunx
encoding.
Converting an OB Van to 3D
If you have access to your truck and some time and money to spare, converting
an OB truck to 3D is a relatively easy task. You need to do the following:
n Bring in 3D monitors with a bare minimum of two, one for program and
one for preview.
n Spend some time in the patch area, duplicating all links.
n Reprogram the video switcher for 3D processing, that is, running the two
High Definition streams in parallel.
n Reprogram the EVSes to process the dual streams.
n Upgrade the DVR or DDR to dual stream capability, as with the Sony
SRW-5,000.
106 3D TV and 3D Cinema
Graphics in 3D
Using 3D graphics may seem to be a no-brainer, for all TV shows are now com-
plemented with full CG graphics that are based on 3D models, and could very
easily be rendered on two channels and included in 3D inside the volumetric
live picture. However, in feature animation postproduction, this is among the
harder jobs done today—so there’s no reason the TV crews would have it easy.
For one, it’s really easy to render CG 3D; the hard part is ensuring that it makes
sense with your real, live picture. Your eyes need both image geometries to be
perfectly stable in regards to each other, and your brain needs both 3D uni-
verses to be coherent in regards to the optical world you’re used to living in.
This second requirement is not an easy one at all.
Consider a situation where we have a stereoscopic problem and a 3D problem.
The stereoscopic problem is that you need to be able to look at the two left and
right pictures, and the 3D problem is that you want the two worlds—the CGI
and the live action—to be plausible when you look at them.
position, pure logic dictates that that person should occlude the graphics.
Because this is not the case with the current generation of CG titles, you’ll have
an unpleasant 3D experience.
In 2D TV, the safe area convention was agreed upon to ensure that no graphic
or action would be displayed in the unsafe borders of the picture. That could
always be cut out by various under-scan processes. In 3D, the same concept can
be extrapolated into safe volumes for graphics.
Have some consideration for the members of your audience who will see the
show in public places, like pubs. They will watch it on passive televisions, with
half the vertical resolution. Some sort of anti-aliasing pass will be welcomed.
For the members of the audience enjoying the show in active stereo, beware
of the effect of display delay on moving logos. Because one eye sees an image
with a slight delay, it is interpreted at a different depth. Depending upon which
image, left or right, is shown first, the movement will be perceived differently:
a title moving leftward or rightward will be perceived either closer or farther
away than expected. On which side of the window will the title be pushed?
There’s no way to predict it until the frame order is standardized in 3D TV.
When you insert a logo, branding, or advertisement for another show, be care-
ful not to do it on a 3D scene that reaches out of the screen. An obvious solu-
tion is to place such images in front of the screen plane, which looks good and
will provide wow 3D. Still, until you can rely on embedded depth metadata,
you have to figure out how much you should pull them.
Another interesting stereoscopic endeavor will be the picture in picture journey
into 3D. If you need to merge a flat source inside a 3D feed—typically a dis-
tant reporter who has no 3D camera—surrounding his image by a frame that
is pulled forward by a few pixels of parallax will go a long way in creating 3D
illusion. If you have a stereoscopic source from a distant studio, give it a key-in
try; you should be able to pull a nice holographic videophone effect.
The Vizrt Stereoscopic SolutionAll 3D TV projects are Synchronization Requirements When ESPN and BSkyB
high-profile productions. They work one HD-SDI chan- started producing S3D content, they used their existing
nel for each left and right camera, all the way through graphics. This made a lot of sense, because everything
the vision mixer, and then either apply a multiplexer to was already CG-3D.
carry the images in side by side, or offer a top-and-bot-
tom frame-compatible format for distribution. For us, that meant doubling the rendering speed. Quad
buffer stereo would be too slow, because it could only
When using the vision mixer, we ought to deliver two deliver half the performance that people are used to.
separate fill and key signals for the left and right eye. Therefore we doubled the rendering stages (as shown in
Because everything we’ve ever done since the beginning Figure 4.7).
of Vizrt has always been openGL-based, it was already
3D. The evolution from a 2D product into a 3D stereo- It is extremely important that these two instances are
scopic operation was very simple. 100% in sync, because if one starts the animation
Shooting 3D for Broadcast or Editing Chapter 4 109
2 x HD SDI
2 x HD SDI
STEREO HD SDI
GPI
2 x HD SDI
STEREO CONTROL
FIGURE 4.6
Stereoscopic Graphic in OB Van Environment. The vision mixer gets two HD-SDI from each camera. The graphics engine receives
GPI signaling and sends back two fill and key HD-SDI.
FIGURE 4.7
Stereoscopic Graphic System. The control application sends animation orders to the stereo controller that distributes the two
rendering GPUs. A synchronization engine ensures the genlock and frame-lock of the S3D fill and key signal sent to the production
matrix via four HD-SDI links.
110 3D TV and 3D Cinema
earlier or later, it’s very disturbing to the human brain. the clock. However, in stereo production, your eyes can-
We created a stereo distributor, which receives the com- not follow the usual flow. If you look for the score and
mands that are usually meant for a single engine and the clock, you would just go to the upper-right corner
distributes them equally to the left and right engines. We from wherever you’re looking. Let’s say you’re looking at
also have a synchronization processor, which ensures the ball right now, and then you move to the upper-right
that both GPUs are always rendering the exact same corner to see how many minutes are left. When you do
field. Those are the two mechanisms making sure anima- that while viewing a stereo image, your eyes would con-
tions start at the same time, and that once the anima- verge on the playing field, then probably on the grand-
tions are running, the fields are always rendered for the stand, and then they’d have to converge again to see
exact same timestamp on the left and the right eye. the score and clock that sits up at the window—which is
a very troublesome experience. To avoid this, you should
Setting Graphics in Depth In the Vizrt engine, we have
virtually push back the 3D score-and-clock element 150
three independent graphics layers: back, middle, and
meters, so it’s near the grandstand. This makes the
front. The back layer is very rarely used in sports pro-
viewing experience a lot more comfortable, because you
duction, where most of the stereo productions are hap-
can follow whatever is in the scene to find your way to
pening right now. Usually the middle layer is used to
the information you need.
show lower-third graphics, like tables, analytical graph-
ics, and starting grids. The front layer is used for boxes, However, this positioning may not work if you go to
such as scoring or clock boxes. Each layer can have stereo rig number two. Let’s say, for example, that the
its own camera, and therefore can have its own stereo steady-cam is located at the border of the playing field,
parameters. Even if we don’t get the best information and its convergence parameters are set so that the play-
from the rigs or from the pixel processors, we can cor- ers get out of the screen as they come closer. (This is
rect the cameras. fairly common.) In this case, what you should do is take
The stereographer usually goes through all cameras the score all the way out and let it sit in front of the
prior to the production, adjusting the stereo parameters monitor by half a meter. You’d do the same, with a more
for the three layers for every single camera. When the subtle change, on the lower third.
director selects, say, camera one, on the vision switcher, You can also disable the front layer so that the back
we get a signal indicating that we’re now looking at the does not get displayed when you use a camera that is
video that is filmed by rig number one. We get that infor- very close to the ground, which allows you to avoid hav-
mation via GPI, and render the graphics according to ing the graphics puncture the players, or any subject on
that camera’s settings. the field. This is currently the common practice.
On the main layer, we would have the lower-third graph-
Adapting Existing Graphics for Stereo You may have
ics. Most of the time these would sit just a bit in front of
a client who gets into S3D and wants to use whatever
the screen, and normally they’re short. Now, let’s say the
footage they had before in stereoscopic production. In
stereo rig number one is the main camera in the sports
order to do this, they must check the design to see if
production; it is over the grandstands, looking onto the
it will generate mistakes. For example, a common mis-
playing fields and over to the other grandstands (a com-
take is to misplace objects and then rescale them. Any
mon scenario). This would give us quite a bit of room
object that is 100 meters away but is scaled up by a
for the lower third to be placed in front, or almost at the
factor of 5 will have the same screen size as an object
level of the screen.
that is very close to you but scaled down by 0.5. If you
People are used to seeing the clock in the upper-right look at this in stereoscopy, it looks completely wrong.
corner: that’s where they would look for the score and This kind of thing must be tested beforehand, such as
Shooting 3D for Broadcast or Editing Chapter 4 111
on the graphics station, where the user interface is 2D camera tracking for a classic camera. If you’re using
but the render context is in full openGL quad-buffering infrared tracking or motion analysis, position the target
3D. You can also check it in stereo while you correct so that the center of the virtual camera is in between
what has been done wrong. In addition, you can see if the two real cameras. There are then many parameters
any of the auto-animations are touching the borders, inside the software that are used to calibrate a studio.
making the viewing experience unpleasant. To that end, You can apply rotations and translations around all
the control PC uses a S3D display and a CG-3D mouse three axes to match the cameras to the target, and with
with six degrees of freedom to adjust position and orien- those parameters you can make sure that the virtual
tation of the graphic elements. cameras for the left eye and the right eye match the
positions of the real cameras. Overall, it’s a very com-
Virtual Studios in 3D The interaxial and convergence plex setup.
parameters in a studio do not really change during a
production, because it’s a very confined space. When Faking the 3D Camera in a Stereoscopic Virtual
you shoot internally, you’re not re-focusing dynamically; Studio If you want to use your existing virtual studio
you basically set it up once and then let it go. You can infrastructure and you only have one anchor presenting,
adjust parameters such as eye separation and con- there is a very simple method of generating a 3D ste-
vergence dynamically to match the position of the pre- reoscopic virtual studio: by feeding the single video into
senter against the virtual studio. the Vizrt engines. Then you offset it in horizontal direc-
tion, matching the distance from the anchor to the virtual
Calibrating Cameras for Stereoscopic Virtual Studios camera.
Studio calibration is a two-step process, and you cali-
brate the lenses as an entity. Basically, you establish You can do that and keep the anchor moving around in
the following: a closed environment, less than one meter back and
forth. Alternatively, you can measure the position of the
n The relation between the zoom and focus on the anchor by installing surveillance cameras looking down
lens. from the lighting grid. This will give the depth position of
n The field of few of the virtual camera in X and Y. the anchor.
n The center shift of the lens.
n The deformation parameter of K1 and K2 for this The really interesting part about this is that you can
lens. make the anchor stand at the exact location that you
want him or her to stand—but what’s even more inter-
Lens distortions are especially important between the esting is that, for most people, the anchor does not
right and left eyes. In order to be very accurate, you need appear flat. This is because the human brain takes a lot
to distort the image according to the distortion of the more hints when it tries to reconstruct a 3D image than
lens. If you film a checkerboard, for example, you will see just the separation of the left and right eyes.
that a wide shot results in the lines at the edges being
curved. To fix this, you should render the virtual studio When we did this experiment the first time, the subject
with a 20% larger size, and then use this result as a tex- standing inside the green screen environment didn’t
ture and re-render it on a grid that has been distorted know about the technology at all. Because she knew it
to replicate the lens aberrations. This ensures that you was in 3D, she reached out her hand to create a wow
don’t see all the artifacts that occur in a distorted image effect. The audience, expecting her hand to come out of
when you try to create two pixels out of one. the screen, actually moved back. This happened in a few
seconds only, and she quickly brought her hand back in.
Tracking Cameras for Stereoscopic Virtual Studios The audience didn’t know whether the hand was reach-
Stereoscopic camera tracking is exactly the same as ing out or not, but they reacted as if they did.
112 3D TV and 3D Cinema
FIGURE 4.8
Using a 2D Camera in a 3D Virtual Studio. If the 2D camera is appropriately placed in depth using X-shift, the presenter is
perceived in 3D.
Shooting 3D for Broadcast or Editing Chapter 4 113
RECORDING 3D
Most 3D takes are recorded for post-edition or postproduction, rather than
aired live. Just as in 2D, the recorded quality and resulting amount of data is
set upon the eventual use of the pictures.
The procedures in stereoscopic recording vary depending upon the final prod-
uct. For a heavy-FX feature, you’ll want to keep all available metadata and
ancillary data you may get. This can include depth information from image
analyzers, rig geometry from motion-control commands and readouts, and
clean plates and helper images. Images intended for heavy VFX treatment
will be recorded at full color resolution with no compression or lossless
compression.
Footage destined to be directly cut into an indie movie or TV documentary can
be recorded in compressed 4:2:2 format. This will maximize recording dura-
tion, allowing for the use of solid state drives rather than fragile magnetic tapes
or drives.
In between, for feature documentaries or low-budget, low FX movies, there
are many workflows based on the raw recording of single-sensor Bayer-filter
cameras.
Hands-on Experience With the Cinedeck light. It does file naming, and you even have a
Bruce walked on stage at the 2010 NAB 3D conference “play 3D” button; there’s no other proprietary
holding the camera unit he had been using just one recorder on the Red or the Phantom that gives
week earlier in the Amazonian jungle, two Silicon Imaging you that. That’s the reason why SDI output is
SI-2Ks on an Element Technica Neutron rig with the crucial on 3D gear: until you can feed the two
Cinedeck recorder. Here are his comments on the system: image streams into any 3D-aware piece of gear,
you are just producing a sluggish pair of flat con-
The Cinedeck is one of the most amazing pieces tent. It’s only when you have a 3D recorder, moni-
of gear I’ve had my hands on lately. We were in tor, or multiplexer that you start doing 3D. And
a place that was 40 degrees Celsius, with 100% remember to only put on tape that you know is
humidity and rain, dust, and wind—but the gear good stereo; otherwise, it goes in the rough edit
still worked. You can see the 3D on-screen, even in 2D, and has to be fixed in post or re-edit, both
if it’s not bright enough to do anaglyph in broad of which will cause great frustration.
Recording Metadata
Metadata recording is a very important issue in 3D, but not one with a solu-
tion. Vendors’ solutions still rule the show, with no standardization or even
coordination announced yet. As of today, rig geometry metadata is recorded
with dedicated devices, on independent supports like USB thumb drives.
Considering all the efforts the industry has been through to record left and
right streams for consistency and coherency, it seems clear that metadata
should be recorded along with the image. The closest we get now is to record it
in one audio channel.
There are three types of metadata available on set to be recorded:
n Orders sent to the cameras and rig concerning zoom, focus, interaxial, and
convergence.
n Values read from the cameras and rig, like interaxial and convergence from
motion-control encoders, as well as values read from the lenses with digital
encoders like Cooke 5/i.
n Computation results from image analyzers, like suggested geometric
corrections.
Shooting 3D for Broadcast or Editing Chapter 4 115
Examples of 3D Recorders
Recording 3D requires twice the bandwidth of 2D. Solid state disk (SSD)
recorders offer such high data throughput that they can record 3D natively. It
is just as easy and transparent to the system to increase the number of hard
disk drive (HDD) disks. Such disks can be stacked up to aggregate their band-
width. On a tape recorder, one needs to increase the tape speed or recording
density, which is not something you can do by buying equipment at your regu-
lar computer shop. For these reasons, most 3D recorders are either SSD record-
ers or high-end HDD recorders using RAID subsystems, with only one 3D tape
recorder in the high-end range.
On location, we use flash drives for their shock, head, and extreme tempera-
ture resistance. Because of their limited capacity (about 40 minutes), they are
handled like film rolls, with a dedicated technician in charge of the backup
and recycling procedures. This includes creating two copies of the footage and
sometimes a Frame Compatible 3D proxy. The recycling can be a simple erase,
but some generations of SSD need to be reformatted to keep their optimal per-
formance level.
On set or in studio, we use RAID disks for their large capacity. If the disks have
to be moved, it is advised that you select a device that has shock-absorbing trays.
You may be aware that disk failures are not uncommon; on high-end systems,
ultra-high value data, such as your latest take, is actually intelligently spread over
many disks, so that a single failure would not compromise the record.
When it comes to backing up and storing data, nothing beats the cost and den-
sity of tape support. If you plan to shoot for hours and hours, don’t rely on
drives: you’ll soon need a truck to move them.
SOFTWARE 3D RECORDERS
Using specialized acquisition software, you can build your own 3D recorder,
or buy it all configured from the integrators. Most of them run on a desktop
and record full-frame 3D via AJA Kona 3 or Blackmagic DeckLink cars. Some
of them can run on a laptop, and will record FC-3D via a Matrox MOX2 or
Blackmagic HDLink encoders. Note that you’ll need an FC-3D multiplexer to
convert the 3D format.
n SiliconDVR (runs on Microsoft Windows): http://www.siliconimaging.com/
n SpecSoft Rave (runs on Linux Ubuntu): http://www.spectsoft.com/
n QtakeHD (runs on Apple OSsX): http://www.qtakehd.com/
There are several types of ingest hardware:
n AJA Kona 3: http://www.aja.com
n Matrox MXO2 Mini and LE: http://www.matrox.com/
n Blackmagic Design: http://www.blackmagic-design.com
FIGURE 4.9
Qtake Software Stereoscopic Recorder.
network operation center (NOC) as a video device, and also serves as a stereoscopic
multiplexer for monitoring on set. The codex recorders are used in top-level pro-
ductions, where they interface with rendering farms and data centers to generate
deliverables. Here are some examples of tape and HDD recorders:
n Codex portable and studio recorders: http://www.codexdigital.com/
stereoscopic/
n SpecSoft RaveDuo HDD recorder: http://www.spectsoft.com/
n Sony SRW-5,000 and SRW-1 CineAlta tape recorders: http://pro.sony.com/
Editing 3D
There are three types of products available for 3D editing:
n 3D plug-ins for 2D edit suites (like the CineForm codec or Stereo3D Toolbox)
n 3D editing suites (like Avid or Sony Vegas)
n Digital intermediate suites (like Assimilate’s Scratch, Iridas’ Speedgrade, or
Quantel’s Pablo)
All these tools take care of the four most important tasks of 3D editing:
1. The coherent ingest of the left and right footage into the edit station.
2. The coherent processing of the left and right footage in the edit.
3. The immediate visualization in 3D of the edit decisions.
4. The rendering of transitions and deliverables in real-time batch processing.
Your requirements for the edit suite reactivity and final render’s image quality
will drive you towards one solution or another, budget permitting.
The CineForm Neo3D codec interfaces itself between the assets, the displays,
and the editing software. When stereoscopic footage is ingested, Neo3D links
them together and presents only a single 2D image to the editing suite. That
2D image can be a flat-eye or a frame-compatible 3D, depending upon your
selection on a small on-screen controller. If you use a dual-stream-capable
118 3D TV and 3D Cinema
hardware card like the AJA Kona 3, the codec will make sure it plays full-frame
3D. What’s important to remember is that the edit suite does not have to be
aware that it actually manipulates stereoscopic content.
The left and right assets are linked together with metadata, called live metadata,
because they describe operations on the images that are applied in real time.
This is more than classic 2D operations like color correction or color look; 3D
operations like image flip and flop, geometric corrections, frame synchroniza-
tion, and depth balancing are also possible. These operations are controlled
from a helper application, CineForm’s First Light, which can access the footage
from a remote station to fix shots while they are edited.
Transitions and plug-ins can be written for the Neo3D system, and will access
both streams of the 3D footage. This allows for 3D-specific image manipula-
tions, like depth transitions or depth effects.
If your edit system of choice is Apple Final Cut Pro, you may be concerned that,
so far, there’s no support for 3D in any Apple product. Hopefully, you can plug in
the Stereo3D Roolbox plug-in that retrofits FCP into a 3D edit suite. This follows
the same philosophy as Neo3D, with a simpler approach and a lighter price tag.
Sony has a consumer-level edit suite, Vegas Pro, which has been upgraded with
stereoscopic functions in its version 10. Sony’s presentation at the SMPTE confer-
ence on 3D production suggested that version 11 will be released with upcoming
Sony 3D camcorders, and that it will have 3D balancing and viewpoint interpo-
lation functions. We’ll see at NAB 2011 whether this is really the case.
FIGURE 4.10
The Stereo3D Toolbox.
Image courtesy of Tim
Dashwood.
Shooting 3D for Broadcast or Editing Chapter 4 119
FIGURE 4.11
PFMatchit Stereo Node.
Image courtesy of Pixel
Farm.
All the Avid edit systems can handle stereoscopic content. The high-end suites
will ingest and play both streams at once, whereas the smaller nonlinear editors
(NLE) will rely on frame-compatible proxies to do “off-line stereoscopic” edits.
In 2002, Iridas created its first stereoscopic applications, and eventually applied
dual-stream technology to its entire catalog. SpeedGrade and FrameCycler are
now totally stereoscopic.
Tracking 3D Cameras
Tracking stereoscopic cameras in 3D is not an easy task, because it requires
much greater accuracy than in 2D. Simon Gibson is the creator of PFTrack and
PFMatchit, two recent and already renowned 3D tracking tools from the Pixel
Factory. Here’s his expert opinion on tracking in 3D:
What specifically makes PFTrack and PFMatchit of both cameras at the same time, and can ensure that
3D-capable? parameters like the interocular distance are constant
throughout the shot, which is unlikely to happen if the
It is the stereo camera model. This ensures that the
two cameras are tracked independently.
same tracking points are used to calculate the position
120 3D TV and 3D Cinema
The left and right eye views of the scene can be tracked for the entire shot. This is a similar issue to tracking
independently of each other, but because of noise and shots where the focal length changes.
other errors, the two cameras obtained may not exactly
Is there anything to take care of on set to help stereo
match the actual camera positions in the stereo rig
3D track?
when the relationship between them is ignored. Tracking
a stereo camera can ensure that the interocular dis- Just measure and record as much data as possible
tance is held constant (assuming this is required). regarding the parameters of the stereo rig and the posi-
tion of items on set. If all else fails, having a set of sur-
The second virtual camera can be generated syntheti-
veyed tracker positions available will help out, as with
cally by translating the position of the first, but this
monoscopic camera tracking.
becomes problematic if the convergence distance is
changing throughout the shot, because the second cam- Do you track the two cameras independently, or do you
era would need to be animated by hand to match the share some of the computation parameters like feature
change in convergence. points identification or matching pair confirmation?
Also, for monoscopic shots captured from a fixed loca- We provide mechanisms for both, mainly because
tion in space (e.g., a tripod), the distance of each track- no one approach is going to work in every situation.
ing point from the camera cannot be estimated due to a Tracking the cameras independently raises issues such
lack of parallax in the shot. Because there is always par- as keeping the distance between the cameras fixed,
allax between the left and right eyes of a stereoscopic or ensuring that the convergence distance changes
shot, the 3D position of each tracking point can still be smoothly throughout the shot if it is not locked off.
estimated. Tracking points that are identified automatically are
Finally, tracking both cameras at the same time using a also transferred from one eye to the other automati-
stereoscopic camera model ensures that they appear at cally. For manually placed tracking points, the position
the correct scale and orientation relative to each other in the other eye can either be specified automatically or
in the virtual scene. If each camera was tracked inde- manually.
pendently without the use of survey data, one is likely to Do you generate some sort of depth map after the
need re-positioning manually to bring it into the correct reconstruction?
position relative to the other.
Not from stereo cameras, although this is something we
Is it easier to track parallel shots than converged 3D? are looking into.
That depends on whether the distance of the conver- What do you foresee in the field in terms of future
gence point from the cameras is known. If not, then tools and future procedures?
this value must be estimated by the solver, which intro-
duces another “unknown” into the system. Tracking with From a tracking point of view, we will be trying to support
a known, fixed, and finite convergence distance is not whatever the production community is using to shoot
really any different from tracking parallel cameras, where stereo. Having a secondary stereoscopic camera on set
the convergence distance is known to be infinite. may become more common from a VFX point of view—
even if the final show is not stereoscopic—because the
If the convergence distance is changing, the tracking stereo camera allows a wider range of data to be cap-
problem becomes more complicated; now there is an tured for various VFX tasks, such as motion capture and
extra unknown for each frame, rather than one unknown set reconstruction.
Dual HD-SDI or
Shoot 3D Dual Gig-E
1 Beyond Wrangler
Record Mini 3D DDR / Viewer
1 Beyond
GoHDMag CineForm
3D Interlaced
1 Beyond
Duplicate / Wrangler Dude
Verify 3D Duplication
GoHDMag Safe
DVI or
GoHDMag GoHDMag
3D Review HD-SDI
or
GoHDCart
1 Beyond Wrangler Pro
3D Review
Avid 3D Edit
FIGURE 4.12
Synoptic of a Stereoscopic Postproduction Workflow.
Image courtesy of 1 Beyond.
your images, that’s most likely to composite them with CG elements. You have
two options: one is to correct the image geometry, and apply CG images on
warped live footage. You’ll need a good tracker to do this. The second option is
to match the CG to the imperfect live image, and correct the composited shot
based on the analysis of the pre-comp imagery.
122 3D TV and 3D Cinema
3D Conversion
3D conversion is gaining a lot of coverage and controversy in the entertain-
ment press, and most commentators do not seem to understand what they’re
talking about. The industry paved the way to this sea of nonsense by endors-
ing ludicrous real-time 3D conversion in 3D TV sets, or daring to release
3D-butchered movies.
Conversion is not a good or a bad thing by itself; it is a visual effects tool—
nothing more, nothing less. The quality of a visual effect is a direct function of
the time and money you put into it. As Jeffrey Katzenberg put it, “Quality is in
the painter, not in the paintbrush.”
In the following years, there’ll be a demand and an offer for bad 3D conver-
sion; distributors will profit on 3D saving a lame movie, some TV channels will
be in desperate need of 3D content, and documentaries will need cheap 3D
stock shoots. The sharks are in the water, but serious producers and filmmakers
will do their best to stay out of the fray.
On the other end of the spectrum, specialized post houses can deliver such
quality conversions that it takes a trained eye to identify them. They’ll work
with directors who consider conversion to be part of their creative work, and
not an afterthought.
In short, conversion as a tool is okay, but conversion as a means is not.
Here are a few key points you should keep in mind:
n Forget about universal solutions for real-time conversion. This is science
fiction.
n 2D/3D conversion is a full FX job and requires time and money.
n Solutions with some “babysitting” may be okay.
Formatting 3D
There is a saying about distributing 3D: “Resolution, synchronization, and
cost: Pick two.” As we know, synchronization is not negotiable in 3D; a lost
frame is a lost 3D broadcast. And because sync is the costliest to maintain, res-
olution is what is commonly left aside in this equation.
COMPRESSING 3D
Before we get into the intricacies of formatting and packaging 3D, let’s say a
word about compression. All 3D requires more bandwidth. This is obvious
for full-frame 3D, where one sends two pictures instead of one, but it’s even
more true for frame-compatible 3D. Anamorphic compression of the images
increases the density of information—in other words, the frequencies. Each
macrobloc in a FC3D hosts twice as much information as its counterpart in
2D. Just to get to the same level of quantization (geek slang for washing out
pictures), you’ll need to allocate more bit stream at the encoder’s output.
There are two additional factors. First, compression artifacts adversely affect
image matching in the visual cortex. Second, lost definition in the blurred areas
124 3D TV and 3D Cinema
eliminates matching features, and echoes on sharp transitions create retinal rival-
ries. This applies to full resolution 3D too. Here again, 3D is more than twice
2D. At contribution levels, it is advised to allocate 20Mb to each 3D channel; at
the final stage, 15Mb is required for a frame-compatible 3D channel on cable.
Most MPEG encoders can process 3D content with a slight software upgrade.
Beware of frame mis-sync between channels, especially with free-running encod-
ers. Adaptive group of pictures (GOP) can get them into incoherent I-frames.
LETTER BOXING 3D
In frame-compatible formats, letter boxing has to be equally applied on left-eye
and right-eye images. On a top-and-bottom image, the total black area is sliced
into ¼, ½, and ¼ parts, as shown in Figure 4.13. Otherwise, the rebuilt images
will not overlap properly on the 3D displays.
glimpse of what will soon be the modus operandi, when 3D content starts to be
produced and distributed by new 3D channel operators on various continents.
3D FORMATS IN CONTRIBUTION
You don’t really know how many times your program will have been re-encoded
and re-framed by the time it reaches its audience. Conversions between
top-and-bottom or side-by-side, 720 or 1080, interlaced or progressive, will add
up with all the frame rate interpolations occurring in international events NOC
and eventually at the display level. The perfect catastrophic combination will
bring you such a damaged 3D picture that it’ll make analog NTSC look good.
The longer you keep your 3D in pristine full frames, the better. Working with
two streams that cannot drift by one frame is not easy, mostly because it was not
a compulsory issue until now.
As a result, the current state of the art is to preserve the 3D at its full resolution
for as long as you control the quality of the signal, and then fall back on frame-
compatible 3D. Satellite links are the usual threshold. During the summer of
2010, the FIFA World Cup 3D signal was sent as separate left and right signals in
JGPEG 2,000, from stadiums to the International Broadcast Centre (IBC), and
then uplinked as side-by-side frame-compatible 3D in MPEG2 HDTV.
ESPN 3D sends its 3D as full-frame 3D from the event venues to its network
operations center in Bristol, Connecticut, and then encodes it in frame-
compatible 3D for the distribution networks.
RESYNCHRONIZING 3D
Stream synchronization is easy to control at the source, but tricky to maintain
along the chain. If one device experiences a glitch, or an overload, it will cause one
frame to get out of sync, and you’ll have to chase down the bug. Experience shows
that sounds tend to get out of sync faster than picture, especially with all the addi-
tional image-processing stages in a typical 3D setup from the camera to your TV.
For example, the consumer industry standard HDMI 1.4 includes a sound delay
flag to take care of the additional processing time for 3D images in the displays.
In most cases, time code (TC) sync is not that helpful, because image delay can
be less than a frame. Sometimes just a few lines of drift will cause a frame drop
later in the line. During transport, one can set a coherent TC in left and right
feeds above the link, and retime it at the receiving station, but that may fail if
someone on the line messes up the TCs. Even worse, if the synchronization
was wrong to begin with, you’ll propagate the imperfection along the line.
Evertz Microsystems has developed a 3D synchronization system that relies
on correlation to correct 3D streams. Their system compares the left and right
frames and searches for the higher correlation. Experience has shown that this
method works for fixing 3D sync. There is always more correlation between left
and right views than with the previous or following frames (with the excep-
tion of panning shots in the direction and speed equivalent to the inter-ocular,
which is a well-known effect in 3D conversion).
126 3D TV and 3D Cinema
revenues are from DVD sales. All of a sudden, the 2.5x revenue gain of 3D the-
ater exhibition turns into a mere 25% increase. Small independent 3D film-
makers have realized that the battle for scarce 3D screens is a tough one, and
that there’s a need for a direct-to-video channel in 3D too.
As of today, there are actually two 3D Blu-ray standards: official and informal.
The official 3D-BD (3D Blu-ray Disc) is a new format designed and promoted
by the Blu-ray Disc Association. It requires 3D encoders and 3D authoring,
and it delivers full-frame 3D. The informal 3D format is frame-compatible 3D,
obviously: side-by-side or top-and-bottom footage recorded on a regular 2D
disk. If you go that route, pay attention to designing the menus in 3D too, so
that the navigation does not requires switching on and off the 3D mode on
the TV. While you’re at it, make both left and right occurrences of the buttons
active, so that the user doesn’t have to guess which one should be clicked. Sony
Vegas Pro10 can export your 3D edits on SBS Blu-rays.
The following are full 3D-BD encoders and authoring suites:
n NetBlender Encoder: http://www.netblender.com/main/
n Sony Blu-print: http://www.sonycreativesoftware.com/bluprint
n Main Concept MVC Encoder: http://www.mainconcept.com/products/
sdks/3d/blu-ray-3dmvc-sdk.html
In addition to these, Roxio is also working on a 3D version of its tools.
INTERNET DELIVERY
On the Internet, 3D video is spreading like wildfire. TubePlus has had a 3D
player since 2009; Netflix is offering 3D video on demand; Next3D is a dedi-
cated service; and Sony PlayStation will also soon have a 3D channel. Samsung
has announced it will distribute 3D content, and Nvidia already does it, with
live events broadcasted on the Internet.
Underground distribution has already started, with FTP services in northern
Europe that host hundreds of gigabytes of 3D content of all levels of legality.
The 3D medium is the perfect content for IP delivery, with a low audience den-
sity that is highly technophile and connected. YouTube 3D is now part of the
Google video monetization service. (What will be the first viral 3D video to hit
a million views? Our bet is a 3D lolcat of Puss in Boots.)
Our point is: more than anywhere, the medium of 3D will not be fighting for
its spot on the Internet, because it is part of geek culture.
Both Roxio and Sony consumer 2D editing tools offer the option to upload
videos to YouTube directly from the timeline. How long before one can upload
full-resolution 3D trailers to Internet servers from an Avid?
Theatrical Distribution
DIGITAL
Digital 3D film distribution relies on the DCI specification. 3D DCPs are
frame-alternative 48fps Jpeg-2000 image streams. The encoding service is
128 3D TV and 3D Cinema
129
LOW-FX 3D FEATURE
Derrière les Murs
Céline Tricart, Binocle, France
3D TV and 3D Cinema.
© 2012 Elsevier Inc. All rights reserved.
130 3D TV and 3D Cinema
FIGURE 5.2
Shooting Derrière les
Murs. Two Genesis
cameras on a Binocle
Brigger III 3D rig.
the camera position is usually the last decision made, so the crew was
basically waiting for us to be ready. That’s the same experience as being
the FX supervisor on a shoot. You are the last to set up before the shoot
rolls on, and you feel like everybody wonders what you are doing.
We eventually timed ourselves, and we were able to change lenses
in as little as 7 minutes; however, with long focal, it was much more
complex, and our average time was closer to 20 minutes. The most
complex rig manipulation was low mode, where the reflective camera
looks down the mirror; we needed that configuration to get the camera
very close to the ground, down to 5 cm rather that 30 cm. (You want to
avoid this for normal shots, because the mirror catches more dust.)
Turning the whole rig upside down and recalibrating the lenses took
us 15 to 30 minutes, depending upon the focal length. Do that twice
to get back to normal configuration, and it’s an average of 40 minutes
of lost time. However, we managed to get these shots first thing in the
morning or last thing at night, so that the time waste was halved.
What was your camera unit composed of?
We shot with Panavision Genesis, Primo lenses, on a Binocle Brigger III.
We recorded most shots on the Panavision Solid State Recorder, the SSR-
1 flash disk. They are great for 3D. On the docking station, it can record
25 minutes of 4:2:2 3D before you send it to be copied onto tape or disk.
There were also a few green screen shots, which we recorded dual
stream 4:4:4 on Codex. This piece is enormous: one cubic meter, noisy,
Stereoscopic Experience from the Front Lines Chapter 5 131
and heavy. Still you need one to record hours of uncompressed full
resolution at full color depth.
As for monitoring, we used Binocle’s Disparity Tagger, which would
warn us of any image symmetry issues. It was on a rolling table, with a
22-inch 3D display used by the stereographer and director to discuss
the depth settings. We frequently used the “final corrected frame”
onscreen display, which shows the eventual image area when it is shot
(which needs to be stereo balanced in post). Additionally, the split-
screen mode and the wiggle 3D preview modes were very useful in
detecting specular lights, which are always troublesome in 3D.
FIGURE 5.3
The Binocle Tagger
image processor on
location.
FIGURE 5.4
Shooting Derrière les Murs.
In late 2008 Jordi joined Mediapro, where he co-leads 3D initiatives with the
research department, as a senior researcher. With his team he participates in
the European research projects IP-Racine, 2020 3D Media, Apidis, Project FINE,
and the Spanish projects i3media and Immersive TV, among others.
Since 2009, Jordi has also been involved as 3D technology consultant and
producer in the 3D live productions of the MediaPro group, with technology
partner Kronomav and Portugal-based company Medialuso. His main tasks are
the development of live 3D tools, 3D workflow, 3D OB van design, and crew
training.
FIGURE 5.6
Shooting RIP Curl World Cup of Surfing, in Peniche, Portugal.
We had the main HD OB van, the 3D OB van, and the satellite link van.
The 3D camera fleet included:
n Five positions with Grass Valley LDK 8000 Elite (ten cameras)
n One position with Toshiba’s 3 CCD Mini (two cameras)
n One Panasonic 3D camera
n All the support staff for engineering and communications were from
Medialuso
136 3D TV and 3D Cinema
FIGURE 5.9
Side-by-Side Rig on Location. A pair of Sony EX3s shooting the Doc Walker concert in Kelowna, Canada.
Image courtesy of Al Caudullo.
138 3D TV and 3D Cinema
I won’t give too many details, but one tool is an amazing white balance
system for pairing 3D cameras. Two 2D cameras will never have the same
color behavior—not even because of the mirror effect, but because their
internal LCDs do not match. Getting them to color match is an endless
battle, unless you use calibration tools designed for 3D like Tim’s.
Regarding creative stereoscopy:
Tim pulled out great shots with moving backgrounds, even with the locked
convergence. Because the dolly was on a semicircular track, the distance to
the band was steady, and in most cases we didn’t even have to reconverge
in post. The free camera could get backstage and audience shots that
would have been much harder to get with a 3D rig. This run-and-gun camera
will be an amazing tool for 3D productions like documentaries and live TV.
Regarding camera positions:
Based on the metrics the production gave us, we originally planned to
have an AG-3D at the top of the hill, but their distance estimation was
so wrong that the IOD was eventually not adapted and the 3D was too
shallow. We took apart the mirror rig that we planned would go on the
dolly, and placed its two EX3s on a side-by-side mount. This created
great 3D, where the miniaturization of the crowd and stage actually
gave the feeling that the crowd was much larger than it really was.
What about the lighting conditions?
The show started at 8:00 p.m. and ended around midnight. We went
through quite a wide set of light conditions, including the sunset at 9:10
p.m. We actually had a show break right after sunset to change stage
lighting; we had huge blasters that shot light at the audience. We were
afraid they’d mess up the images with flares and all sorts of direct light
issues, but it eventually worked very well, with the flood of light helping
to layer the 3D with all sorts of halo effects on the band and audience.
FIGURE 5.10
The AG3D-A1 on a Dolly. Shooting the Doc Walker concert in Kelowna, Canada. Operator: Tim Dashwood.
Image courtesy of Al Caudullo.
140 3D TV and 3D Cinema
FIGURE 5.11
Doc Walker Wide Shot. Shooting the Doc Walker concert in Kelowna, Canada.
Image courtesy of Al Caudullo.
CHAPTER 6
Opinions on Creative
Stereoscopy
141
RAY ZONE
Ray Zone is a 3D film producer, speaker, and award-winning 3D artist. Zone has
produced or published over 130 3D comic books and is the author of Stereoscopic
Cinema and the Origins of 3D Film: 1838–1952 (University Press of Kentucky, 2007)
and 3D Filmmakers: Conversations with Creators of Stereoscopic Motion Pictures
(Scarecrow Press, 2005). Zone’s website is viewable in anaglyphic 3D at: www
.ray3dzone.com.
FIGURE 6.1
Ray 3D Zone.
In 1933, shortly after the invention and proliferation of sound motion pic-
tures, philosopher and critic Rudolph Arnheim, in Film as Art (University of
3D TV and 3D Cinema.
© 2012 Elsevier Inc. All rights reserved.
142 3D TV and 3D Cinema
California Press, 1957), wrote that “The temptation to increase the size of the
screen goes with the desire for colored, stereoscopic, and sound film. It is
the wish of the people who do not know that artistic effect is bound up with the
limitations of the medium and who want quantity rather than quality. They
want to keep on getting nearer to nature and do not realize that they thereby
make it increasingly difficult for film to be art.”
As an early theorist of cinematic art, Arnheim has isolated the fundamen-
tal artistic problem of the stereoscopic cinema. Very simply, it is this: tech-
nological advances leading to greater realism can create a corresponding loss
of artistry. Raymond Spottiswoode, in A Grammar of the Film (University of
California Press, 1950), shared similar concerns when writing in 1933 about
the “unreality of the film image”; he was “exceedingly wary about the advan-
tages of color (except in animated films) because [he] fears that it will prove
yet another step on the road backward to a mere imitation of life” but noted
that a few exceptional color films had “been able to move in the border world
between abstraction and reality, and so share in the advantages of both.”
Foreshadowing his later work with 3D movies, Spottiswoode also wrote:
If the stereoscopic film were ever realized, it would seem that it too could
enjoy these advantages. There is a world of solid shapes far removed
from the luscious figures and glamorous interiors with which Hollywood
will fill its stereoscopic movies. And beyond the third dimension looms the
fourth. Even within the limits of present space, stereoptics can become a
powerful instrument for transcending reality, not merely imitating it.
A great struggle is now evident in the motion picture community, particularly with
respect to professionals in the fields who write, direct, photograph, and edit the
movies. These professionals have learned to speak a planar visual language. They
have built their professional careers on, and become facile at working with, a funda-
mental defect of cinema. That deficiency is the flatness of the motion picture screen.
Now that digital technology has facilitated the repair of defective cinema, narra-
tive artistry is at a loss. It no longer can readily resort to a familiar tool. A new lan-
guage and new linguistics for z-axis storytelling are necessary. And, ironically, it is
Sergei Eisenstein, one of the chief architects of the use of montage in cinema, who
has perhaps most articulately suggested, even in vague outline, the importance
of this new language. The “entire course of theatrical history,” Eisenstein wrote,
“through the centuries, at practically every step, unfailingly and consistently
reveals the self-same tendency—distinct in its forms, yet single in purpose—to
‘cover’ the breach, to ‘throw a bridge’ across the gulf separating the spectator and
the actor.” In the last essay Eisenstein ever wrote, the master of montage character-
ized stereoscopic cinema as “a dream of unity between spectator and actor.”
The stereoscopic realism now available to the motion picture storyteller can
reinforce this dream of unity in the cinema. And it can drive equally well a nar-
rative in the service of the abstract or the real. What is essential is the growth of
a republic of dreamers equally adept at using the z-axis parameter for abstrac-
tion or realism in telling new stories on the stereoscopic screen.
FIGURE 6.2
The Cinema Screen.
Image courtesy of Ray 3D
Zone.
144 3D TV and 3D Cinema
YVES PUPULIN
Yves Pupulin is among the few movie professionals who had an early interest in the cre-
ative potential of stereoscopic 3D images. He started working on feature films in the
1980s as an AC, and then joined the French company Excalibur, where he supervised visual
effects. As a true French stereoscopic cinema pioneer, Yves Pupulin spent twelve years
developing new tools like dedicated camera motion control. He is one of the inventors
of the first motion control 3D camera, and has worked on human brain and stereoscopy
research. He is also one of the animators of the industry research ad-hoc group 3DLive. In
1998, he founded Binocle, a European leader in stereoscopic 3D movie and TV production.
Binocle was founded by cinematographers and engineers in order to create efficient ste-
reoscopic production tools. This toolset includes 6-axis motorized camera rigs and soft-
ware suites for real-time analysis and processing of 3D feeds. The Disparity Tagger is a
GPU-based 3D monitoring and correction system dedicated to 3D TV and 3D cinema pro-
ductions. The Disparity Killer is a software-only solution designed for 3D postproduction.
FIGURE 6.3
Yves Pupulin, Binocle Chairman.
The following are reflections on the status of stereoscopy today, with some spe-
cific focus on its mise-en-scène. These reflections emerged out of various meet-
ings and conferences on the subject, and our intent is to clarify in which ways
this technique is linked to others and to foresee its development from an aes-
thetic, historical, and economic standpoint. Can stereoscopy find a place within
art history? Will it develop in parallel to cinema, or will it acquire its own mise-
en-scène? Does it require a specific crew for its completion? Finally, we will con-
sider stereoscopy in its current context, in order to envision its prospects.
Opinions on Creative Stereoscopy Chapter 6 145
Stereoscopy: A 3D Cinematography?
While the search for volumetric and spatial presentation punctuates art his-
tory, the conditions for the existence of a stereoscopic art form remain unclear.
When we speak of stereoscopy today, we don’t quite know whether it consists
of an evolution of cinema, or of a cinema of its own kind. For this reason, we
will begin by examining which distant or recent art forms can shed light on the
relationship between stereoscopy and 2D.
Everyone knows cinema does not need volumetric presentation in order to exist.
Fihman and Eyzikman, known for their work on holography, defined it as an
industrial art, in the footsteps of Walter Benjamin. Let us consider the industrial arts
simply as those requiring a technical device to enable their distribution on a great
scale. This is the case for cinema, radio, and video. In order to define the specifics of
stereoscopy, we ought to ask ourselves what particular status stereoscopy and holog-
raphy can hold among them, as well as what relationship they have to one another
and to technically more distant visual art forms such as painting and sculpture.
FIGURE 6.5A
The Thinker. Example of haut-relief sculpture, tympanum of The Gates of Hell, Auguste Rodin.
Rodin Museum, Paris. Photography by Richard Kirsch.
FIGURE 6.5B
Man Falling. Example of haut-relief sculpture, lower-left part of the tympanum, The Gates of Hell,
Auguste Rodin.
Rodin Museum, Paris. Photography by Richard Kirsch.
Opinions on Creative Stereoscopy Chapter 6 147
FIGURE 6.6
The Monument of the Burghers of Calais. Example of monumental sculpture, Auguste Rodin.
Rodin Museum, Paris. Photography by Richard Kirsch.
In stereoscopy, the technical devices and audience system are very close to
those of cinema; this is because, in both cases, a screen in a dark room is being
viewed. This is why stereoscopy is very often considered the natural evolution
of 2D cinema; the new stage in cinematic techniques.
But does this perception, which seems to recognize the value of stereoscopy,
give it a chance to exist viably and to become, like 2D cinema, an arena for a
mise-en-scène constantly renewed, a practice reinvented each moment, on every
set, in every editing room, in every auditorium, or on every color timing device?
A comparison with holography sheds light on this question. For when we
speak of 3D, do we really speak of stereoscopy? Or is it rather, unconsciously,
148 3D TV and 3D Cinema
Sculpture can be divided into two main categories: low and high relief on the
one hand, and monumental sculpture on the other. If a parallel can be made
between sculpture and 3D, we will suggest that low and high reliefs are more
closely related to stereoscopy and monumental sculpture to holography.
Indeed, painting and low and high relief can only be observed from an angle infe-
rior to 180 degrees, just as in cinema and stereoscopy. Moreover, certain condi-
tions exist for their visibility (framing, architecture, projection screen) and part of
the piece remains “invisible” to the viewers (related to the backside of its support).
In a sense, the stereoscopic image is similar to the 21⁄2D image (with depth)
described by neurobiologist David Marr as an intermediate neuronal stage
between the 2D retina and the 3D interpretation that reaches our conscience.
Claude Baiblé, who teaches stereoscopy to aspiring cinematographers at the
Louis Lumière National School in Paris, notices that images, whether in 2D or
21⁄2D, remain attached to their original point of view. The stereoscopic image
does copy 3D reality but does not equate it: the viewer can move his head,
stretch his neck, and the visual will remain the same, which is of course not the
case in direct vision where each time the point of view is moved, a new appear-
ance is created—a new glimpse—instantly convertible into O3D reality, allo-
centric, or centered on the object rather than the point of view.
The passage from 2D1⁄2 to 3D (from “egocentric view field” to “reality before
us”) indeed implies that we access—by sliding or moving the eyes—a volu-
metric vision of objects (O3D) independent of the successive images: “what
I am seeing is reality spatialized (I3D englobing O3D) and not a semitrans-
parent hologram (I3D) or a stereoscopic image (I2D1⁄2) attached to my cyclo-
pean eye.” In cinema, moving the head has no effect on the image obtained;
the image cannot, therefore, go beyond the stage of 2D1⁄2 representation, and is
besides fully dependent on a point of view previously imposed.
For more on this subject, readers fluent in French can refer to Claude Bailblé,
“En relief, l’image et le son,” Cahier Louis-Lumière 4 (June 2007): .
the invention of cinema that this problem has arisen. We have been able to
make 3D films for a century, thanks to the anaglyphic projection of left and right
images onto one sole support; however, anaglyphs are not satisfying for the brain,
because of their drawbacks: loss of color, ghosting, and great visual strain. Starting
in the 1950s, synchronized double projection and polarization to separate the
eyes partially solved the issue; still, it was impossible to produce stereoscopic cou-
ples without disparities and with a lack of fixity for the brain. Stereoscopy had to
wait for the advent of digital in order to develop, because of the ability to system-
atically correct images pixel by pixel and of the now perfect fixity of projection.
In this stage of stereoscopy’s history, we notice its ambivalence. It falls on an aes-
thetic and technical plane in the continuity of cinema, but it is not yet an art on
its own. There is no such thing as a purely stereoscopic aesthetic; indeed, we con-
stantly use a glossary and notions borrowed from cinema to approach stereoscopy.
Nonetheless, it creates a physiological rupture, just as the differences between
photography and cinema did. Here, the reproduction of depth of varying
geometries reinvents an interpretation of space, imagined by the director.
Now, as far as the process of making a film, how must we think of this ambiv-
alence? Does the physiological rupture also imply a rupture in the mise-en-
scène practices? Is there continuity between the processes of shooting in 2D
and in stereoscopy?
There are technical parameters specific to the stereoscopic mise-en-scène: conver-
gence and interaxial. Today, we have tools at our disposal that enable us to man-
age and master these specifically stereoscopic parameters. Yet, only experience
allows us to anticipate the limits of using 2D mise-en-scène elements inside a 3D
production. This might constrain or restrain the possibilities of reusing your 2D
mise-en-scène palette of tools.
The problem is the emergence of visual discomfort for the viewer—this is what
will limit the combined use of parameters from 2D and 3D mise-en-scène. Digital
techniques make it possible to avoid or to correct some of these disagreements.
By working on the combination of these parameters, within the limits inherent to
the physiological specificity of stereoscopy, an artistic expression specific to stereos-
copy can emerge. However, for this to happen, it will be necessary to identify the
given limits in order to determine, for each 2D and 3D technical parameter and
their combined use, the possibilities offered to the stereoscopic mise-en-scène.
POINT OF VIEW
Cinema remains first and foremost a matter of point of view: the place from which
we observe the object we are about to frame, light, and film. It is the point from
which the director wishes the viewer to observe the world, and specifically the
scene that is taking place before the camera. When a movement of the camera
is involved, the evolution of the point of view makes sense, but also adds, by the
variation of surfaces of lights and shadows, a graphic rhythm to the 2D image.
During the location scouting or preparation of a 3D film, it is necessary to have
a viewfinder in order to locate the different axes and focal lengths, just as in
2D. To scope out the volumetric landscape itself is also important, and at this
stage requires no other tool than visual observation. For this, we need to con-
verge on the position of the subject, which occurs naturally, and at the same
time on infinity, in order to scope out the succession of volumes in depth and
their evolution in the simulation of the camera movement we wish to obtain.
This scoping of volumes is crucial.
Volume can often be better appreciated during movement, when we are close
to the subject, and with a rather short focal length. But it is difficult to make a
film without close-ups; therefore, the scoping out of the set’s volumes becomes
decisive for the placement of the actors, and the point of view becomes deci-
sive for the distribution of volumes in depth and along the edges of the frame.
In holography, we see an object reconstituted with light that can be observed
from all sides. The viewer’s place becomes indifferent, and the holographic
filmmaker will no longer assign one single point of view. Therefore, because of
this viewer’s place, the audience system of cinema itself is questioned, or per-
haps abandoned.
THE FRAME
The precision of the cinematic frame is one of the essential parameters of
shooting. Its rules, whether they are bypassed or not, are very precise, and often
come from painting. Throughout the centuries, painting evolved from a subject
described in its entirety or a portrait as the identification of a character, to the
cutting of the body into fragments evoking a suggestion—a play with the frame
and the out-of-frame.
Opinions on Creative Stereoscopy Chapter 6 153
The frame is what the painter, and now the director, gives the audience to see.
Its exactness in 2D cinema is one of the fundamental parameters of the image.
It can escape no one, be they professionals or viewers; it cannot be crowded by
elements that do not belong in it, and its value is always intentional.
In stereoscopy, the frame is imprecise.
Even though stereoscopy is projected onto a 2D screen, the precision of
the frame is lost in comparison to 2D, because of the zones of disparity
around its depth created by binocular shooting allowing stereoscopic vision.
Anything entering the field on the edge of the frame is visible by one eye
before the other. This is generally irrelevant if the volumetry is soft and the
subject is in a zone with moderate contrast. This visual discrepancy is not
easily quantifiable today and is the result of an accumulation of parameters
that will make it acceptable or not. In all cases, just as in low and high relief,
the edge of the frame becomes imprecise; for elements to pop out requires
the use of floating windows. In bas-relief, objects on the edge of the frame—
in this case, cast into the material, as seen on many historical buildings—
present various details in depth that can even go outside of the sculpted
frame. There is no way to solve this imprecision, for if the sculptor were to
place a frame inside of the first frame, he or she would still face the indeci-
sion of the edges of this included frame.
The stereoscopic frame presents analogous imprecision. When the sub-
ject stands in the depth of the image and a rectilinear element stands on the
edge of the projection window, the frame can give an illusion of perfection.
However, this imprecision, rather than being detrimental, can become one of
the elements of stereoscopic mise-en-scène; a high-angle zoom on a landscape
of hills or on the bark of a tree can give an idea of the variations enclosed
within the projection frame itself. Finally, the imprecision of the frame is dif-
ferent from the variation of the frame linked to image correction.
This imprecise value of the stereoscopic frame pushes some stereographers
towards conceiving an evermore immersive experience for the viewer (for
example, by enlarging the projection screen in order to push back the perceiv-
able limits of the frame).
As for holographic technique, the frame from the screen has disappeared. We
are seeing a subject created thanks to structured light, and, just as in monu-
mental sculpture, we can therefore, in theory, walk around it. There is no lon-
ger a hidden side. We could imagine a decoupage in terms of frames, we could
easily picture a bust placed in the middle of the audience—but can the out-of-
frame still exist?
There might be two types of holography: one that could be observed from the
outside, just as in theater or the circus, and the other from the inside, in an
immersive space, just as if we were placed inside a snowball, which is done in
certain installations.
154 3D TV and 3D Cinema
LIGHT
In cinema, the subject, as a lit object, can receive any contrast and any of its
temporal modulations, only limited by cinematographic support. It consists of,
on the one hand, the source of white light, and on the other, the complete
shutting off of this source, which the viewer can read as the contrast ratio of
the screen. The emulsion, either as a support almost entirely transparent or as
a black image, causing a near complete obturation, has been used in all eras of
cinema—to suggest a storm, for example, or the repeated explosions of a battle
or a fire. While there is a blinding effect that forbids the lengthy repetition of
these successions of on and off, nothing on a physiological level will cause any
damage as long as the norms for lighting and contrast on projection screens
are respected. The blinding effect is part of the mise-en-scène, and its purpose
is to provide emotion, not to damage the viewer’s vision.
In 2D cinematography, light and contrast are only limited by projection norms
and the power of the projector. However, the case is very different for stere-
oscopy. In 3D, high-contrast shots, particularly towards the edges, through
foliage, via backlighting, or amongst the different components of the sub-
ject, can trigger a real cerebral discomfort for the viewer. This is also linked to
stereoscopic occlusion, due to two separate points of view and to ghosting.
This occlusion is the result of the contrast on the filmed subject, of the depth
bracket, and of the projection device on the viewer’s brain. To gauge them is
even trickier, as other phenomena can further complicate this analysis, such as
the polarizing effect of the beam-splitter mirror or the light interaction with
different surfaces.
Lowering the contrast is therefore advised, though we cannot yet today quan-
tify by how much. Our tools must evolve from shooting to projection. During
shooting, the sensors must cover a greater dynamic in order to obtain images
able to maintain volume in highlights and shadows. These sensors will enable
a better adjustment of contrast during the color timing of stereoscopic mov-
ies. The material for perceiving 3D must also progress, the glasses in particular.
Manufacturers are working on this.
In holography, structured light seems to emanate from the subject itself, and
there is no longer a projection frame. The holographic reproduction of a sub-
ject might reach a realism of shapes and matter that we can hardly imagine
today.
EDITING
In cinema, the shortest clip is only limited by the length of projection of one
image; in television, it is limited by the display of the interlaced field. The lon-
gest clip, on the other hand, is limited by the duration of the program or of the
film itself, such as in Hitchcock’s Rope (1948), which was actually constructed
of a few shots with quasi-indiscernible splicing, or in Russian Ark (2002),
which Aleksandr Sokurov filmed in a single 96-minute sequence shot, at the
State Hermitage Museum in St. Petersburg.
Opinions on Creative Stereoscopy Chapter 6 155
than the discontinuity that was expected with going from stereoscopy to 2D
then back to stereoscopy. This phenomenon is even more striking on shots
with depth continuity.
During the capture of a Julien Clerc concert in 2009, a 2D camera framing an
extreme wide shot of the audience and the stage revealed the skyline of Lyon,
and the continuity effect made it evident that it was within the brain’s capacity
to create a 3D vision from a 2D shot, edited between stereoscopic images. This
effect is not perceived by all individuals, independently of their status as pro-
fessionals or audience members.
Similar observations had been made by viewers of Hitchcock’s 3D film Dial M
for Murder (1954), in response to the numerous 2D shots that were inserted
inside of the 3D sequences, some of which remained invisible for some view-
ers. With the use of image-correction software, this phenomenon is even more
pertinent because of the complete absence of accidental disparities between
images, which creates a gentler progression of 2D towards 3D and between the
different 3D shots. In such cases, the brain perceives continuity rather than the
actual split between the two techniques.
To dwell further on these questions, we must ponder the stereoscopic shooting
experience. Dealing with the depth of field, for example, is a very important
component of a 2D image, since it enables a centering of the mise-en-scène
on the subject. This is all the more convincing if the subject is a simple one.
Limiting the depth of field on a filmed object, such as the face of an actress, is
often desired for photogenic reasons or to soften the part of the face cut by the
frame, but it is done mainly to enhance the intensity of the acting and of the
eyes. Camera operators therefore add up neutral densities in front of the lenses
in order to limit the depth of field.
The adverse effects of focus blur in stereoscopy have been documented. What is
really the reason for this, other than the fact that it is not desirable to shoot all
films with an 11 aperture? Experience makes certain observations evident: if the
volumetry is soft, the contrast is reasonable in zones of high disparity, and the
viewer’s gaze is not attracted to the blurry zones, then it is perfectly acceptable. On
the close-up of an actress, a sharp volumetry would, of course, not be photogenic.
Binocle did some tests on limiting the depth of field on feature-length trailers
for such films as the forthcoming The Krostons, directed by Frederik Du Chau.
The result proves satisfying up to a certain limit, at which point there is indeed
a discontinuity or disconnect in what the brain can tolerate.
Limiting the depth of field in 2D is not only aesthetic, it is also narrative or
meaningful. In the film Husbands (1970), directed by John Cassavetes, a man
and a woman are sitting on the edge of a bed in a bedroom in London. We
see them full-face. They lie down on the bed one after the other, and meet in
a limited depth of field when they get back up. Naturally, this passage tells of
the relation between the characters, and thus evokes the relationship between
women and men. Limiting the depth of field is also used on exterior shoots,
Opinions on Creative Stereoscopy Chapter 6 157
when one wants to center an action by the depth of field, as in the case of the
face. It is also a matter of avoiding unimportant details that could crowd and
interfere with the shot.
The importance of the use of the technical parameters of 2D as artistic param-
eters is fundamental in 3D. Consider that 2D has a depth bracket reduced to
zero. As a consequence, everything that is valid in 2D can also be valid in ste-
reoscopy, if we follow, step by step, a gentle evolution of the different variables
of shooting. A graphic curve that would represent the viewer’s disconnect or
the end of his or her cerebral comfort will be more or less steep, depending on
the viewer and the parameters that are varied.
The point of disconnect is easily identified when one parameter is studied at a
time, but not so easily when they are put together. Tools for mise-en-scène and
controls such as the Disparity Tagger are, in this respect, indispensable, because
they make it possible to integrate geometrical or contrast limits on images
corrected in real time according to an average audience comfort setting. This
software also allows other limits to be included for each sequence according to
the director’s artistic sensibility.
From a practical point of view, it is essential to preserve a continuity between
shooting practices in 2D and 3D, so that the same footage can be used for
both cases. But different editing practices are required. If the opposite is done,
discontinuity will be frequent and the viewer’s comfort will be affected. This
question is particularly delicate in the shooting of sporting events using a fast
moving dolly or crane, where experience shows that the classical practices of
2D cannot be entirely transposed to stereoscopy.
Stereoscopy Jobs
We have just raised questions about the ambivalence of stereoscopy, which
depends on cinema while also having to break from it and find its own means.
If stereoscopy triggers a reevaluation of mise-en-scène and shooting practices
and the necessity to create postproduction images beyond the editing process,
which new skills will it require from the crew and the technicians involved in
its making and, in the short-term, in its invention? There isn’t a complete dis-
continuity between stereoscopy and 2D, but rather the need for a variation. We
need to test and use all the variables of 2D while being aware of the limits of
brain comfort and of the accumulation effect.
It is also impossible to strip the stereogram from the screen and to transform
it into a hologram. Stereoscopy is included in the realm of cinema, but does
not cover all of the possibilities offered and used by 2D without creating a dis-
connect in what the viewer’s brain can tolerate, relative to stereoscopy’s own
parameters. So the use of technical shooting parameters in a stereoscopic
mise-en-scène lies inside the play within these limits. Its conception is made
by modulating the depth value throughout the process, sequence by sequence,
shot by shot. And the one to give sense to this whole conception is not the
158 3D TV and 3D Cinema
stereographer, but the director. Only the director can think about the volume
throughout the film, depending on the aesthetic, the emotions, and the mean-
ing he or she wishes to convey to the viewer.
It is a commonly accepted concept that a 3D film must be written for stere-
oscopy. This point can be argued if the director is involved in the writing of
the script and is already imagining the principles she will be using during her
shoot; she will organize the succession of sequences according to the stereo-
scopic directing she has in mind.
There is no topic that cannot inspire a 2D movie if the director decides so.
Why would it be any different for stereoscopy? The artistic rendering into ste-
reoscopic form only exists in the director’s mind, in its process of technical
realization by his or her associates.
It is while creating the shooting script and/or the storyboard that indications
relative to stereoscopy come into play. How should we answer the question,
“What denotes whether a screenplay was written for 3D?” which was asked
during a meeting of the CST (Commission Supérieure Technique du Centre
National de la Cinématographie)? The only coherent answer is “because the
director conceived of his movie in 3D.”
The profession most closely resembling that of a stereographer is a special
effects (SFX) supervisor. We can establish a parallel between the two because
of the similarity of their interventions throughout all the stages of production.
The stereographer must know the shooting conditions, the postproduction
operations, and their follow-up in the digital lab, all the way to the projection
of the movie.
Just as with the SFX supervisor, the stereographer is, first and foremost, the
director’s advisor. His role is to get the crews and tools to create the director’s
desired mood on the whole film, up to its finalization. He must therefore lis-
ten to the DP, who remains responsible for the image, and also works with the
head set decorators, costume designers, and all the crew members contributing
to the making of a film.
positions. For this reason, adding effects in post cannot be avoided. Inspecting
the geometry-corrected images must help him to scope out the many traps of this
double image, of its effects on the edge of the frame, of its contrast, and more.
Meanwhile, the camera operator and grip have to address the cerebral stress
related to the speed of a movement subject to stroboscopy, or by an actor
entering the frame and violating the stereoscopic window. They must also visu-
alize the frame after volume correction, not before. The head decorator can
no longer use trompe l’oeil, false perspective, or backdrops, without wonder-
ing about their stereoscopic value and the revealing of depth that will ensue.
Make-up artists, hair stylists, and costume designers must be careful with
the products they use. In other words: all the professions are involved in this
rethinking; every technician must become a stereographer in his or her domain
of expertise. Finally, when moving throughout the set’s depth, the actors must
comprehend the volume in which they move about, and be aware of where
their gaze falls, taking into account both points of view.
Eventually, the stereographer will be less needed to train everyone on the set,
but she will still remain the one who manages the stereographer’s crew and
ensures continuity of volumetry on the film in a relationship with the entire
crew working on a production.
During postproduction, the stereographer supervises correction, the position-
ing of the window for sequences that will be shot in parallel (oftentimes shots
with special effects), floating windows, and the homogenization of the entire
film; for example, by creating dissolves not of shots, but of depth value.
Final Thoughts
In the 1980s, there were predictions that a wide-scale colorization of cinematic
works shot in black and white would take place, along with the invention of
automatic systems of television colorization. Fortunately, this did not hap-
pen. But this concept of technical progress may be more easily realized with
stereoscopic conversion—with the aid of a Zmap, for example—than its col-
oring counterpart. This could be useful for scientific applications, but by no
means for cinema or television, where there is the desire to cater to an audi-
ence whom we expect to be demanding on an artistic level.
Today, the demands of new and competing markets are what is carrying the devel-
opment of 3D cinema, accelerating a switch to digital screening rooms and ensur-
ing new technology to television viewers as well as the users of new media. Of
course, there is still the possibility that this momentum could be stopped by new
perspectives or by the satisfaction of having reached a necessary and sufficient stage.
The deployment of Digital Cinema projectors in the theaters, justified by 3D,
enabled the digitalization of movies and therefore eliminated transportation needs;
it also enabled the protection of digital copies, thanks to watermarking. For televi-
sion, the technological capacity to automatically turn a 2D image into 3D may seem
satisfying, but it will have as a consequence the progressive destruction of 2D TV,
162 3D TV and 3D Cinema
and, in same the process the abandonment of a true reflection on the artistic pos-
sibilities of a creative stereoscopy. The stereoscopic revolution may then experience a
sudden death both in television and in cinema, its two main sources of revenue.
For now, the recent success of stereoscopy is undeniable and increasing, even
though 3D is often criticized as being too geared towards spectacle. We may
Opinions on Creative Stereoscopy Chapter 6 163
hope, then, that the dream of stereographers and audiences alike also encom-
passes the tremendous possibility of a renewed art medium and refreshed nar-
ratives, capable, like other art forms, of triggering new ideas.
STEVE SCHKLAIR
Steve Schklair has been working at the front edge of new technologies for most of his
career, and has left a mark in movies, special effects, and interactive media. Steve
is currently focused on the development and production of new digital S3D motion
picture technologies and the real-time broadcast of S3D programming. He is highly
esteemed by an international client list as one of the world’s leading experts in digital
and live-action S3D and is one of the primary catalysts behind the recent resurgence
of S3D in Hollywood films.
Steve served as 3D and digital image producer on 3ality Digital’s U23D movie. In
addition to supervising the production, Steve also oversaw the postproduction and
completion at the 3ality Digital studio in Burbank. Other recent projects include the
first live S3D broadcasts of a full NFL game delivered via satellite to theaters on
both the east and west coasts, and the training and technology for the launch of the
BSkyB S3D channel. Prior to 3ality, Steve was a senior executive at Digital Domain,
the special effects studio responsible for films such as Apollo 13, The Fifth Element,
Titanic, and Terminator 2: 3D, creative director for R/Greenberg Associates, and
executive producer of The Columbus Project for computer graphics and interactive
media pioneer Robert Abel.
Steve is a frequent speaker on S3D and new entertainment technologies at leading
international digital and entertainment technology events. He is an award-winning
member of the Visual Effects Society (VES), an associate member of the American
Society of Cinematographers (ASC), and an alumnus of the University of Southern
California (USC) School of Cinematic Arts master’s program.
FIGURE 6.7
Steve Schklair, Founder and CEO, 3ality Digital Systems.
164 3D TV and 3D Cinema
Inception of Modern 3D
I started thinking about 3D camera development in the middle of the 1990s,
when the FCC mandated digital television. At the time, I foresaw that the
best use of digital television was stereoscopic presentation, but that there had
to be a path to create content on a realistic production schedule with realis-
tic tools, and, to some extent, in real time (because television is a broadcast
medium). There were no 3D televisions in the middle to late 1990s, and there
really weren’t any in the early 2000s; thus, even though I was building gear
that would work for television, it also had to work for feature films. This was
always the plan. Anything that works for television is capable of working in
real time—which means it works even better for features because you don’t
have the editing timeline in a real-time broadcast.
FIGURE 6.8
The Arri Depth Chart.
task, and, in those days, we didn’t have tools (such as image overlays) that told
us where we were. We’d line up one camera on a chart, and then we’d make
sure the other camera also lined up. By today’s standards, it was very crude—
not even close in alignment.
possible. That’s how we came up with the function of tracking zoom lenses,
which was critical.
BALANCING COLOR
We performed color balancing via a nine-by-nine color matrix that was incred-
ibly sophisticated, as opposed to three primaries. Because of this, we were able
to remove all the aberrations that a beam-splitter puts in—keeping in mind that
you can’t take everything out, because then you lose some of your depth clues. It’s
important to keep some of the aberrations, because they’re critical to depth per-
ception; depth isn’t judged just on the basis of convergence and interaxial, but on
everything else as well: color, shape, receding sizes, occlusion, and more (there are
around 20 or 30 depth clues in 2D). We didn’t want to lose any of the key depth
clues, so we focused on removing geometric aberrations, pretty much everything.
Another problem we wanted to address was the possibility of someone jump-
ing in front of one camera and not the other. To avoid this, we built a system
that would constantly look at the stereo images. If the image was the kind
of thing that would hurt to look at, like someone blocking one eye and not
the other, the image processor would detect it within the space of one frame,
and would switch to a 2D version from the good eye until the blockage was
over. Then it would switch back to 3D. In most cases, you can’t even detect the
switch, because it took the good eye and did a little bit of an offset to it and
put that straight up in the screen. And it’s all happening in real time.
3D OB VANS
There’s not a lot of difference between a 2D OB truck and a 3D OB truck. Both
have facilities for laying graphics, audio, engineering, and recording, and a pro-
duction area where your director, graphics people, and producers sit. There’s
some difference in terms of equipment and monitoring, but there are more
similarities than differences—and when you really get down to the basics, the
only difference is 3D monitors. (We’ll get to that in a moment.)
To make a truck 3D, all we had to do was add image processors so that we
could have a stereo view of the cameras, and additional image-processing
channels so that we could do graphics offsets. With controllable offsets, we can
steal a shot from 2D cameras. With more image controllers, the instant replay
170 3D TV and 3D Cinema
guys can all get a quick preview of what they are looking at in 3D. Because
they’re cutting, you never know what they’re cutting together, and you don’t
want a lot of depth jumps in the material. At the end of the day, though, it’s
not that different—except for the monitoring.
The monitoring is significant to a few specific people. First, the convergence
pullers who look at the technical monitors. In our broadcasts, they are all look-
ing at a subtractive view, also known as a difference view. Above that is a 3D
monitor showing the line cut, which is more of a confidence monitor to check
how the shots actually look in stereo.
Second, technicians in charge of replays need to have monitors so they can
watch replay packages. In football, for example, a replay will often have a
package built and ready within seconds of the play ending; that’s how fast the
equipment builds these quick edits. Naturally, it’s better if the replay techni-
cians can look at it once in 3D, to avoid airing anything that has problems.
Third, the director needs to see the program and the preview in 3D. (He
doesn’t need to look at all the small camera monitors in 3D; he can figure out
what those look like.)
hit your marks or you don’t. There’s creativity in the director of photography
deciding where the focus should be, but that’s separate from the actual act of
turning the knob to keep things in focus.
If you know the rules (for example, each shot is 2% of background parallax
at the farthest object, and 2% of negative parallax at the nearest object, with
no edge violation), you don’t need to know anything about stereo at all to sit
there with an interaxial and convergence control. You can get anybody, even
people who have never picked up a controller; they can set up an interaxial
convergence. They just read the words, pick up the dial, and go straight to
that knob, that setting. They will have to play a little until they find it. So they
would be balancing the interaxial convergence settings until they have that
measurable setting.
There’s no reason why image-recognition technology can’t do these camera
adjustments when all you’re doing is trying to match the measured settings.
The same is true, to some extent, for the job of the stereographer; there’s no
reason why the stereographer can’t define the depth of the space and let the
software adjust the cameras to meet her needs.
Eventually, I believe machines and image processing can take over all non-
creative work. But machines are not capable of anything creative; only people
can do that. And 3D should remain a creative medium.
MONITORING 3D
Shooting stereo used to be considered to be voodoo and black magic, because
you were never able to see anything until one or two days after shooting, when
you could see the dailies. Establishing camera settings was almost a black art,
involving magic calculations that were known only to a very few. Now, how-
ever, almost anyone can make a good stereo shot if he or she knows how to
use the tools, whether it be a 50/50 mix, a flicker view, a checkerboard, or a
subtractive.
To be clear, I’m not saying that you should be wearing 3D glasses; your 3D
monitor is only for confidence. In my opinion, the best way to get a 3D shot
is not to wear glasses and instead to look at my negative and positive image
parallax. This data is independent of whether someone’s eyes are tired or not,
172 3D TV and 3D Cinema
You don’t have to blow up your picture, you don’t have to compromise your
data, and you don’t have to waste a lot of time. You might want to spend some
time adjusting your 3D in postproduction, and you’ll certainly need to do
depth balancing at the edit, but at least you’re done shooting with 80% to 85%
of your work done.
But, again, in an effects movie, you’re opening up every shot anyway, so
shooting parallel provides a consistency that simplifies the effects process.
Ultimately, you end up with a converged movie, but at least you are giving the
effects people a little more room to work.
With the new tools that are now available, we can actually see 3D on a set. In
some cases, this may compromise the depth when you shoot parallel. In prac-
tice everyone is staring at the 3D monitor, and when you are shooting parallel
everything is negative. You’ve got a director and clients all staring at that moni-
tor asking, “Why isn’t this shot comfortable?” So the stereographer is going to
make a few adjustments to make it a little more comfortable. All he’s doing,
because you shoot parallel, is closing up the interaxial, which is going to pres-
sure depth. You are going to end up compromising your depth when every-
body says the shot is not comfortable or doesn’t look good.
DEPTH TRANSITIONS
When working on the U23D show, the director wanted both a lot of cuts and
also a lot of multiple elements layering. However, multiple element layering
using the material as we shot it would have been visual chaos—so we built a
tool that let us very quickly reconverge shots. Nowadays, most tools do that.
But it let us do it fast enough so that it became part of our production tool.
What we would do was reconverge shots so that wide shots had more positive
parallax, while close-ups were more negative; in other words, we wanted the
images in wide shots to be farther away, and close-ups to be closer. This was
a general rebalance that we did on this film, especially in terms of layers. But
in terms of every set, the most important thing this new tool offered was the
opportunity to do this adjustment in the edit.
The technique we used was to put the monitors into a flicker mode, and then
overlay the two shots. We’d adjust them over the space of 24 frames to 36
frames, depending on how large the move was, so that a shot that was playing
with the last second or second and a half would start a depth adjustment. This
would meet the incoming shot at some halfway point, and then the incoming
shot would continue the adjustment until it was back to where the filmmaker
had intended. We were basically doing soft transitions of depth in about two
seconds; and wOr whatever you edit maybe, one second in front or one second
behind the cut, if there wasn’t a lot of jump. When there wasn’t a wide shot to
wide shot issue, the transitions could be much faster.
The fast cutting worked. Without jumping at depth, fast cuts work. You just
can’t jump at depth. It’s so easy that there’s not much to say about it. This stuff
is incredibly easy to do.
174 3D TV and 3D Cinema
as you’re mixing image size and field of view, narrowing it up without confin-
ing the character, the space is closing up as well. That’s kind of cool.
There are so many ways you can use depth to help analyze and tell a story or
to help an audience feel, to foreshadow, and we used all these things for fore-
shadowing. You can foreshadow things about to happen by adding depth to
the tools you’re already using.
GOING FURTHER
The language of 3D is one that is not yet developed—everything we discuss
here is still in the theory phase. I’ve done some experimentation with 3D sym-
bolism, but it takes an extraordinary amount of preparation time, which most
projects can’t afford. Still, we get a little more of it on every project that comes
up. And as directors of photography start to learn this language, then you’ll
see more and more of it because it will become second nature. That’s the most
exciting thing about previews: it’s a whole other tool in the box and lets us
redefine what the language of film is.
experience? These questions are still being explored, and the answers are still
being learned.
VINCE PACE
As an acclaimed cinematographer and co-inventor of the world’s most advanced ste-
reoscopic 3D system, known as Fusion 3D, Vince Pace is leading the effort to rein-
vent the entertainment experience. He has been instrumental in capturing some of
the world’s most captivating productions of our time. His career marks many signifi-
cant achievements within the industry, from supporting the camera challenge of BBC’s
award-winning series Blue Planet to designing innovative lighting technology for The
Abyss and being nominated for an Emmy in Cinematography for bringing back images
from the deck of the Bismarck, three miles under the ocean. His contribution to enter-
tainment has been in the areas of natural history, features, sports, and concerts.
Almost a decade ago, Academy Award-winning director James Cameron, president of
PACE Patrick Campbell, and PACE founder Vince Pace shared a desire to revolutionize
the entertainment experience. Together, they embarked on a world’s first, to develop
innovative technology for the most advanced stereoscopic acquisition system ever cre-
ated, known today as Fusion 3D.
Today, PACE/Cameron’s Fusion 3D system is trusted by many of the top directors and
entertainment companies in the industry. The use of PACE’s innovative technology has
resulted in over $4 billion of box office receipts and major 3D benchmarks recognized
within the industry, including Avatar; TRON: Legacy; Resident Evil: Afterlife; U23D; ESPN
Masters, NBA Finals, and All-Star Games; Journey to the Center of the Earth; Hannah
Montana/Miley Cyrus: Best of Both Worlds Concert Film; and The Final Destination.
Based in Burbank, California, Vince Pace’s company PACE houses a full creative and
technical team that infuses “Imagination through Innovation” into major motion pic-
tures, live sports and entertainment experiences, concerts, music videos, and more.
FIGURE 6.9
Vince Pace, Founder, CEO, PaceHD.
178 3D TV and 3D Cinema
Many people make the mistake of assuming that we previously didn’t interpret
depth and dimension when looking at 2D images, and that 3D changed that.
In fact, however, cinematographers have long been dealing with perspective
and techniques that enhance the view of depth dimension in the 2D frame.
This new venture into 3D shows the growth of professional individuals who
have been doing successful work in 2D—whether it’s storytelling, composition,
lighting, or cinematography—and who are now learning about the additional
benefit of working in the 3D medium.
As opposed to an almost entirely new language that we need to learn, I feel
that 3D is just a better methodology for conveying an artist’s work to the
viewer. For me, it’s been exciting to work with some of the best in the 2D busi-
ness, and to see them embrace what 3D is adding to their ability to convey a
message and apply their own creative talents. Again, though, it’s important to
realize that we did not invent depth and dimension overnight; it has always
been a part of the cinematographer’s toolset. It’s just that now we are empha-
sizing it more in the way we present it to the viewer.
Live production presents the same challenges as feature productions. In major
feature productions, they spend so much money that you don’t have a second
chance to shoot anything twice. In live production, the ability for the equip-
ment to perform correctly the first time is even more critical; all the money in
the world won’t buy that shoot again.
Unfortunately, most productions realize the hard way that there’s a lot more
to the production system than simply a camera. The creative profession-
als involved are also a part of it: the cinematographer, the AC, the editorial
department, the mobile unit companies, and many others. For this reason,
the strength and growth of 3D will come on the back of 2D: those who did
2D well are the prime candidates to do 3D well. There’s no reason to start
over with a whole new set of rules or a whole new set of people. Similarly,
companies like Discovery or ESPN are the perfect ones to create this new
channel or stream of content, as opposed to adding new or different players
to the game.
Obviously, there are many differences between 2D and 3D. What would be a
simple lens change in 2D is a whole configuration change in 3D, with mirror
rigs, side-by-side rigs, inter-ocular control, and so on. The complexity and vari-
ety of configurations is much greater in 3D, and in the future we’ll see more
of a variety of configurations that service a particular need in production. Our
concentration will be on introducing as many different configurations and
variety as the cinematographer needs. It all started with a beam-splitter and a
side-by-side rig—but soon we’ll have aerials, cable cams, handhelds on a foot-
ball field, sky cams over the field, slow-motion cameras, 500 mm long lenses,
and robotics-based cameras.
Opinions on Creative Stereoscopy Chapter 6 179
What has been successful in 2D will be successful in 3D. Image quality still
matters—3D does not mean we no longer have to worry about it. After all, if
we’re going to ask the public to pay more, we have to make sure we deliver.
Some of these solutions work very well in the B-roll environment, when you
need to pick a shot or to send someone out to the locker room. But most of
our concentration is on the real heart of production, on the production cam-
eras that capture the images on the field, and where the audience expects the
quality to be the best that it can be.
Editing Live 3D TV
The key in 3D is balance. That is, you must balance the cameras so that when
you cut from one camera to the other, it’s pleasing to the audience. People
need to refrain from just looking at one particular image or one isolated event;
the real trick to successful 3D is when they all work together. This is similar to
an orchestra, where all the instruments have to complement each other. When
they are taken out of context, it can be disruptive.
The key for anyone in a live production is to look at it as a complete whole.
Too often people get excited about one camera, or the 3D effect of 3D, but
you don’t want the experience of the viewer to be a highlight reel. Some of the
young guys in the business get too excited about that one camera that gives a
strong 3D effect, when, in order to convey the event, you need the cameras to
complement each other.
We need to get beyond the effect of 3D and into the fact that it’s a better view-
ing experience than 2D. The question is “Did you enjoy that more than a 2D
presentation?” If the answer to that is “Yes,” you won. If the answer is “It was
really cool when this and that happened, but I was missing the score, I did not
get the action on the field,” then you kinda lost what the production was about.
Camera Positions in 3D
There’s a misconception that all 3D can do is get the camera closer to the
action, but this is incorrect. A 3D experience still has to tell the story. But to do
this, you need premium camera positions—that is, where 2D cameras would
be. You can’t put 3D cameras in a corner. We are working very hard at getting
to the premium positions and delivering a 2D feed as well as a 3D feed. That’s
the key, to get to where the 2D is really told. Sometimes it’s helpful to have
lower camera positions to enhance 3D, but that’s not where the whole 3D pro-
duction value is. I’d rather win the battle of having a key camera position, a
court camera, or a slash camera on a basketball game, rather than being some-
where else on the court and that’s all I have. I want to deliver everything the
viewer wants to see in 2D—but in 3D.
We are certainly learning. The definitions of 3D filming and broadcast are in
their inception. The filmmakers and production people are getting over the
poke-it-in-your-face effect and are starting to realize that a little 3D goes a long
way. The power of 3D is so much greater than people realize that they don’t have
180 3D TV and 3D Cinema
to use it as much. In the future, it’s going to be a more pleasing way to watch
entertainment. In fact, the goal should be to make you forget it’s 3D in the first
five minutes—what I want you to do is to get into the project. If you’re only
thinking about the 3D, then you have left everything else behind. High-level cre-
atives in this business, they get it. They understand that this is a medium that can
help them do their business. It does not have to be all about the 3D.
Monitoring 3D
It’s not a requirement for a director to have a 3D monitor—it’s more important
that the cinematographer has the 3D monitor. That said, all the directors I have
seen working with 3D monitors have used them to their advantage to better set
the performance and the actors. It helps gets everybody on board and excited.
Now, how much time did the directors spend concentrating on that monitor?
Maybe only 3%, because they quickly realize that there’s so much more for
them to get done. The 3D is an end result, not a creative result.
For the same reason, there’s no need to run dailies on a large screen. Good 3D
is scalable. I had firsthand experience with this on Avatar; if it looks good on
my 45-inch monitor, I can guarantee it will look good on my 30-foot screen.
When people say, “I have to see it on a 30-foot monitor to make my decision,”
in my opinion it’s an excuse.
And it’s also not a question of experience with stereoscopy; you judge 3D by
the visual cues that are necessary. 3D a DP he can’t judge lighting on a small
monitor. The visual cues for 3D, just as for correct lighting, are visible on a
small monitor as well as on a big screen. To reiterate: good 3D, if you know the
visual cues, is very scalable.
TECHNICAL
Properly producing and displaying 3D used to be a technical nightmare, but
recently turned into a mere digital challenge. Any geek with a sweet tooth for
cable mazes and overheating power supplies had a blast in the early years of
Opinions on Creative Stereoscopy Chapter 6 181
digital stereoscopy. If you are into motion controls, optics, data wrangling, or
in any technical or engineering job in the entertainment industry, you are most
likely experiencing that 3D challenge in your everyday activity.
Most of this challenge will soon—in, say, a couple years—be behind us, with
the deployment of standardized tools and procedures, intelligent rigs, and top-
notch image analyzers. There will always be a new frontier where new image
acquisition and processing systems are invented. Still, we have a road map of
what the 3D path will eventually look like.
ARTISTIC
The second challenge we face is to learn how to create 3D effects with a pur-
pose. An introductory lecture on stereoscopic 3D production usually ends with
the concept that we are about to reinvent the visual art. I sometimes phrase
it like this: “We are a generation of artists, who have an incredible chance to
rewrite the rules of our art.” Art forms have always been extended by radi-
cal artists who refuse to follow the rules and instead invent new paths. Some
of them revolutionized their art, like the cubist painters or the nouvelle vague
directors. Stereoscopic 3D has the potential to put all entertainment profes-
sionals in such a position. There’s an open debate on the most important rule
of 3D: Is it “Do no harm” or “Break all the rules”? The former is an obvious
statement in the entertainment industry, but the latter is a cornerstone of artis-
tic creativity. Constraints are the number one source for creativity. If Spielberg
had had a shark, we never would have had Jaws.
SCIENTIFIC
The third challenge is more difficult to describe—even to define or name. It
relates to the way we process the stereoscopic 3D imagery, from our eyes to
our feelings, in their biological incarnation somewhere in the wetware we host
between our ears. The challenge is to understand what happens with the eyes
in the brain’s visual cortex when we see natural 3D and when we watch arti-
ficial stereoscopic 3D. There’s consensus that this understanding should be
called “the human factor in 3D.” It covers medical science, neurology, optom-
etry, cognition, evolution, psychology, and numerous other hard and soft sci-
ence domains. It is quite a new research area, not widely addressed until now,
as it is not a public health subject; only highly selected and trained profession-
als, like medical researchers or car designers, use 3D displays in their everyday
life. In the general population, only specialized optometrists were concerned
with the binocular vision of their patients. And because seeing in 3D isn’t
critical to life, stereo-blindness was not considered a disease that deserved a
cure. To our knowledge, the first and only stereopsis therapy documented
so far is “Stereo Sue” in 2007 (see http://www.fixingmygaze.com/ for more
information).
We are not implying that the wide release of stereoscopic 3D imagery is a pub-
lic health hazard that should be legislated by the U.S. Surgeon General or the
182 3D TV and 3D Cinema
U.N. World Health Organization. Far from it, as preliminary research points to
the opposite; S3D seems actually to help in detecting and curing potential stra-
bismus. What we have to address is merely the issue of visual fatigue. What we
want to understand is the interactions between visual comfort and visual activity,
between spatial placements and inner feelings, between motions and reflexes.
You may have heard 3D specialists discard this opinion. To our understanding,
unless you are speaking with one of the top five stereographers in the world,
disregarding such reality is either infatuated ego or unfair marketing. The truth
is, we are still creating the tools and learning how to use them.
Nonrealistic Depth
How great would it be to be able to change one stereoscopic characteristic
without affecting the others? In other words, to be able to inflate or deflate a
character, make him rounder or flatter, farther away . . . and yet not bigger?
Some of this will never be feasible, but a surprising number of visual cheats are
now possible in CGI stereoscopy, and soon in heavy VFXed real images.
The basic effect is the dual-rig rendering, with one interaxial for the back-
ground and another for the foreground. This creates nonrealistic 3D universes
where volumes, positions, and sizes are not related. It’s a form of per-depth-
layer stereoscopic setting that can be achieved with real imagery shot on green
screen. The Coraline (2009) team did a great adaptation of this by building
miniature sets that were compressed along the z-axis. They performed their
magical puppetry and shot what is, to our knowledge, the only stereoscopic
that is both 100% practical effects and compressed depth.
Opinions on Creative Stereoscopy Chapter 6 185
Depth (Dis)continuity
Continuities are all over the cinematographic art. Light, color, frame, action,
movement . . . we constantly take care not to disturb the viewer’s mind beyond
186 3D TV and 3D Cinema
that thin threshold bordering the suspension of disbelief. The same care
should be given to our story’s depth settings; depth continuity is as important
as any other continuity. This means that we want to respect it as much as pos-
sible, for the sake of our stories.
Depth continuity lies both within and across shots, and the amount of depth
discontinuity you impose on images creates a depth velocity. That depth veloc-
ity has to be followed by the viewers’ eyes and brains. This is not like color; it’s
more like sound or light, which can exhaust the audience into discomfort, or
even wake them up from their cinematographic second state.
You’ll manage to keep the stereoscopic velocity under control up to the point
that you have to edit a cut that does not match; these types of cuts are known
as depth jump cuts. The worst example of this kind of cut is going from a far
away landscape to a close-up talking head. Steve Schklair explains his cut-
cushioning technique in his section of this chapter.
Whatever stress you impose on the audience’s visual systems, remember to
provide them with depth rest areas, where their ocular-muscular system is not
called to action. It’s important to remember that 3D visual fatigue is muscular
fatigue caused more by the duration of the effort than its intensity.
LINES OF DEBATE
Good stereographers make use of all existing 3D storytelling tools. That being
said, you will see various emphasis on this or that, depending upon the rela-
tionship between the tool and the stereographer’s overall 3D style or current
interpretation of a story. When used as a key creative element, a tool will define
a style. In other cases, it will be used as a mere technical solution to smooth an
edit locally or straighten the discourse.
Is There a Screen?
As we have seen, the screen disappears from the viewer’s sight and mind
when 3D imagery is adequately performed. Yet the wall at the end of the the-
ater room, where the sound comes from, still exists. The cinema space has a
boundary—and the screen, where the image we are looking at is located, is the
medium our eyes are locked on while we follow the story unfolding in space.
So, is there a screen or not? Yes and no. That’s up to the cinematographer. On
the one hand, there’s a screen, an optical object, that you can’t control and
that has a secondary role; on the other hand, there’s a stereoscopic window, a
strong storytelling object, that you can place and move at will.
Depending upon your willingness to move away from 2D cinematography, and
your confidence in new 3D visual effects tools, you’ll stay in the “screen” para-
digm or jump into the “window” one. No need to rush here.
screen plane. If you remove your 3D glasses, you will definitely see two movies
at once and want to put your glasses on as soon as possible to get back into the
stereoscopic storytelling.
In 2010, Zack Snyder’s Legend of the Guardians: The Owls of Ga’Hoole received
the closest thing to a standing ovation on the specialized mailing list of 3D
cinematographers, in large part for its stereoscopic treatment, which consists
of putting almost all the action behind the screen, up to a sustained parallax of
3%, or 34 pixels—an unheard-of number until then.
Another approach is not to consider where your action is relative to the screen
plane, but to concentrate on the size and volume of the characters. Depending
upon the value on the shot, from close-up to large, this will place the char-
acters anywhere from 1% (25 pixels at 2 K) to 1⁄2% (15 pixels). You’ll
then make extensive use of the floating window and cut cushioning to
accommodate the framing and smooth the cuts. This approach was devel-
oped by “Captain 3D,” Phil McNally of DreamWorks. It requires implement-
ing two postproduction 3D effects directly into the CGI production tool. The
DreamWorks 3D Camera in Maya has both effects implemented and animated
at the 3D layout level, offering live 3D preview in the 3D viewport of the pro-
posed final edit.
In Avatar, the object of attention is always in the screen plane. This is espe-
cially obvious during dialogue shots, when two characters talk alternately. The
convergence point is shifted continuously to keep our eyes converged on the
screen. If you remove your 3D glasses when two protagonists talk, you’ll see
the speaker’s image clear and sharp, and the listener’s fuzzy and duplicated. As
soon as the role shifts, the visual presence does the same.
3D FATIGUE
There are two main sources for 3D fatigue: muscular-ocular and cerebral
effort. The causes of muscular fatigue are the non-canonical use of the visual
muscles: the eye lens and the extra-ocular muscles used to control our sight
direction. The causes of cerebral fatigue are all the discrepancies in the image
domain that impair the surface matching and other factors involved in the
stereoscopic fusion.
between bad 3D sequences. On the other hand, your brain can cope with some
defaults, and will get more and more used to them. Left/right inversion is an
interesting example of this; if you don’t detect it and decide to cope with it,
you’ll eventually forget about it. But you will still get serious brain shear.
The actual subject and content of the 3D images will affect the time factor. Let’s
consider the synchronization default, when left and right images are not dis-
played in the appropriate synchrony. The amount of stereoscopic disparity is a
direct result of the time disparity: no action, no default. However, any horizon-
tal movement generates depth artifacts, and vertical ones generate unpleasant
rivalry. Consider someone walking, for example; his torso will be in a constant
false depth position. His arms and legs will oscillate from the false depth of the
torso at the shoulders and hips, to a back-and-forth exaggerated depth position
at the hands and feet. The standing foot will be in the correct depth, while the
moving one will either pass through the other leg or dig into the ground.
Even if we could rule out technical mistakes, we still have to deal with geomet-
rically correct stereoscopy that still produces potentially disturbing 3D effects,
for example, strong negative parallax, on-your-lap 3D, and other eye-popping
moments. When is 3D too much? There may not be an academic answer
beyond, say, two degrees of divergence—which will eventually be acknowl-
edged as the stereoscopic sibling of the 100 dB sound pressure pain level. Badly
mixed soundtracks insult the audience long before reaching that legal thresh-
old. When is 3D too much?
Stereographers invariably answer such questions evasively. Game and 3D art-
ist Peter Anderson’s catchphrase is said to be “the situation dictates.” My own
experience giving 3D lectures has been that the best answer is that the solu-
tion depends on what you are framing and on how you’ll edit it. After a while,
I’ll also say that it has to do with the music score; if there’s a silence or a loud
explosion, you can and can’t do certain things. Eventually, the attendees will
understand they are asking an artistic question, not a technical one. There’s no
theoretical answer to an artistic question.
We could get away with this if we were only a bunch of creatives involved
in 3D. We all know the extent of the magic tricks a good movie editor, color
timer, or talented soundtrack artist can do to save the day when the material
we are working with is not blending. Still, that’s not a valid answer in oversee-
ing the deployment of a new mass media visual form.
This simplification has been so powerful that we still hear that the cameras and
projectors should be placed 65 mm apart. All these distances are called inter-
ocular distances, even though we actually have an inter-pupillary and an
inter-optical distance. This shows us that the art of 3D started as an art of rep-
licating human vision, not as entertainment—similar to what it would have
been like if painting had been invented by naturalists. As long as most 3D
movies were science movies in IMAX theaters, or illusionist entertainment in
special venues, this made sense.
(FOV) they offer. My cell phone is five times smaller than my laptop, yet offers
50% of its FOV. The TV set in this hotel room is 2.5 times wider than my lap-
top, yet offers 50% of its FOV. That said, this TV is too far away. When I watch
movies on my computer, I push it farther away than when I’m typing, where it
will occupy the same FOV as this TV.
Creating 3D only for specific screen sizes would be comparable to designing
all of our electronic displays, including theaters, for an FOV, not a surface. And
this is actually the case. Still, the field of view tends to increase with screen size,
as we do not want our cell phones to block our view, yet we want the cinema
to be an immersing experience.
Now, remember that the 3D screen is a window. Let’s consider three typical
screen sizes: cinema, TV, and cell phone. Where would you have a 10 m win-
dow, which is basically a glass wall? In a sky resort, a beach house, or any other
gorgeous vista—in other words, somewhere it makes sense. Extra-large win-
dows are for exceptional views. Where would you have a 1 m wide window?
Anywhere you need daylight. A nice view would be great, but not compulsory.
Finally, where would you have a 3-inch window? On a prison cell door to watch
inmates, or on a fortified wall to shoot at enemies; to see without being seen, to
control, to harm. These three windows describe the whole spectrum of feelings
of freedom, from gorgeous vacation resorts to life and death behind bars. How
can you pretend to tell the same story through such different mediums?
It happens to be the same in 3D trigonometry. We have seen how the viewing
distance and screen size interact, but what about the IPD? In a theater, there’s
a point where it becomes negligible, because we have the ability to diverge our
vision to a certain degree—say, one degree—and remember, it’s all about the
music. At some point, that one degree of divergence becomes more important
than the increasingly positive parallax on screen. It all depends on your abil-
ity to diverge, but there’s one ratio (distance-to-the-screen versus screen-size, or
size/distance) where you’ll always be comfortable.
Regarding the two other classes of screens, hand-helds and TVs, we have either
the IPD metrics superseding the screen, or the IPD and screen size in a bal-
ance of power. Just as IPD jumps in at this size/distance, so does the focus/
convergence. Our depth of field (DOF) is proportional to our focusing dis-
tance: the farther away we look at something, the deeper our in-focus box is.
On the other hand, the closer we look at something, the smaller our DOF is.
One can hold a laptop close enough to emulate a theater screen FOV, but it
will not recreate the comfort of a large faraway screen. The same is true in 3D.
One cannot fuse images at stereoscopic infinity (65 mm parallax) on a desktop
screen; they have to be scaled down. Negative parallaxes require the same treat-
ment. To some extent, the reduction of the FOV that comes with the reduction
of the screen size and viewing distance follows the same curve as the reduction
of the dioptric box presented by John Merritt. This has been our great chance as
stereographers: to be able to work on various screen sizes and still be able to
judge the quality of a picture.
3D A to Z
195
Richard W. Kroon
#
2D compatibility The extent to which stereoscopic content can be processed
by standard 2D equipment without a negative impact on the content.
2D for 3D;—capture Applying stereoscopic best practices when shoot-
ing and editing a 2D work to support a possible later 2D to 3D conversion.
Otherwise, the finished work may be difficult to convert to 3D or may deliver
disappointing 3D results.
2D to 3D conversion The process of producing a 3D image from a 2D
source. This is a computationally complex process. Low-quality, inexpensive
conversions have a cardboard cutout feel, with a series of flat objects set at dif-
ferent distances within the scene. Higher quality and considerably more expen-
sive conversions can approach the appearance of works originally recorded
with full stereoscopic views.
3D-compatible An electronic device that can detect and pass stereoscopic
content on to another device without altering or acting upon the stereoscopic
content in any way.
3D display An image delivery system that presents separate views to the left
and right eye, creating a three-dimensional representation. Commercially avail-
able 3D displays can be categorized generally as stereoscopic (requiring special
eyewear) and autostereoscopic (viewable without eyewear).
3D metadata Additional information recorded along with stereograms to
assist in their decoding or presentation.
3D signal processing The analysis and manipulation of electronic data
unique to stereoscopic images. Processing equipment must be aware of the
type of signal it is processing (2D or 3D) so that it can take appropriate steps
to produce the desired result.
3D source master The original material from which a stereoscopic program is
produced. Before being released for digital cinema, the 3D source master is first
assembled into a collection of uncompressed picture, sound, and data elements
(the digital cinema digital master, or DCDM) and then compressed and encrypted
to form the digital cinema package (DCP) for delivery to the theater. Alternatively,
before being released to the home entertainment market, the 3D source master
is first assembled into an uncompressed and unencrypted 3D home master and
then into compressed and possibly encrypted 3D distribution data.
3D TV and 3D Cinema.
© 2012 Elsevier Inc. All rights reserved.
196 3D TV and 3D Cinema
A
accommodation Changes in the shape of the eye’s lens that allow one to
focus on objects at different distances, such as when following a close moving
object. Similar to adjusting the focus on a camera’s lens.
accommodation/vergence; accommodation/convergence;—link;—relationship
The neural association between accommodation (focal adjustments to the eyes)
and vergence (turning the eyes in equal but opposite amounts to line up on the
same point in space). Useful in the estimation of depth for objects up to 200
meters away (stereo infinity). Matching accommodation and vergence signals for
objects closer than stereo infinity reduces eyestrain, improves depth perception,
and decreases the time required to perceive a three-dimensional image.
accommodation/vergence reflex; accommodation/convergence reflex The
automatic focal adjustments made as the eyes converge or diverge. This reflex-
ive response must be overcome when viewing a stereogram, since the focal dis-
tance to the screen is constant even though the amount of vergence changes
when viewing different objects at different apparent distances.
accommodation/vergence rivalry; accommodation/convergence rivalry A
mismatch between the amount of accommodation (actual focal distance)
and vergence (based on apparent object distance) required to clearly view a
three-dimensional presentation. Common when viewing stereoscopic images;
because everything is at the same distance from the viewer (on the surface of
the screen), even different objects appear to be at different distances (thanks to
positive or negative parallax).
active stereo A stereoscopic device that requires the viewer wear active eye-
wear that switches the right and left lenses on and off in time to the presenta-
tion of left- and right-eye images on the display. Autostereoscopic devices and
technologies based on passive eyewear are excluded.
anaglyph; anaglyphic process A frame-compatible stereogram where the
left- and right-eye images are color-coded and combined into a single image.
When the combined image is viewed through glasses with lenses of the cor-
responding colors, a three-dimensional image is perceived. Different com-
plementary color combinations have been used over the years, but the most
common are red and cyan, a shade of blue-green. Anaglyphs may be projected,
displayed on a monitor, or produced in print.
3D A to Z 197
B
barrel distortion A lens aberration where straight lines bow outward away
from the center of the image and rectangles take on the shape of a barrel.
Resulting from a decrease in focal length moving outward from the center of
the lens. Most common in short focal length (wide angle) lenses. The effect
becomes more pronounced as the image aspect ratio increases. The amount of
barrel distortion is generally expressed as a percentage of the undistorted pic-
ture height. Also fisheye effect.
beam-splitter; beam splitter An optical device that divides
incoming light into two parts using prisms or mirrors, such as a
50/50 two-way mirror; a device that is both reflective and trans-
missive, though not necessarily in equal measures.
binocular Relating to having two eyes or coordinating the use of both eyes.
binocular cue; binocular depth cue Depth information that can only be
interpreted by calculating the differences between two simultaneous views of
the same scene taken from different perspectives, generally those of the two
eyes separated by the width of the nose.
binocular depth perception The ability to perceive three-dimensional space
and judge distances based on the differences in the images recorded by the left
and right eyes (disparity) and the angle of the eyes (vergence).
binocular disparity; disparity The differences between the images perceived
by the left and right eyes due to the physical separation of the eyes. Those same
differences are simulated in stereoscopic systems to create the illusion of three-
dimensional depth from otherwise flat images.
198 3D TV and 3D Cinema
binocular rivalry When the differences between left- and right-eye images
are so large that binocular fusion is not possible, resulting in a confused visual
state. The brain may alternate between the two eye images, may select portions
of each eye image and blend them into a sort of variable mosaic, may interpret
the overlaid eye images as static, etc.
binocular suppression When the brain ignores some portion of the
image from one eye and only registers the overlapping image from the
other eye. Occurs in response to diplopia (double vision), since the sup-
pression of one eye image eliminates the double vision that would other-
wise have resulted.
binocular vision Visual perception based on two eyes working together.
Binocular vision provides a number of evolutionary advantages, hence its pop-
ularity within the animal kingdom. The most important feature of binocular
vision for 3D systems is depth perception.
binocular vision disability A defect in the vision system that prevents proper
interpretation of binocular depth information. Approximately 5–10% of the
population suffers from a binocular vision disability sufficient to interfere with
the perception of depth in a stereoscopic presentation.
brain shear The brain’s inability to properly and comfortably fuse poorly-
constructed stereographic imagery into a three-dimensional representation.
Causes eye fatigue, double vision, and, in extreme cases, pain or nausea.
Coined by American film director James Cameron in 2009.
C
camera coordinate system A three-axis (x, y, and z) system of Cartesian coor-
dinates (oriented so x, y, and z represent width, height, and depth,
respectively, and grow in a positive direction to the right, up, and
away, respectively) oriented to a camera’s imaging plane so that the
origin is aligned with the principal point (the point where the optical
axis of the lens intersects the image plane).
cardboarding; cardboard cutout; cardboard effect The lack of a
true 3D effect in a stereogram, giving the appearance that the image
is made up of a set of flat cardboard cutouts as in a child’s pop-up book.
Generally caused by inadequate depth resolution in the image due to a mis-
match between the focal length of the recording lens (or virtual CGI camera)
and the interaxial distance between the cameras. Also a common side effect of
poor-quality 2D to 3D conversion.
checkerboard; CK;—3D A time-sequential stereoscopic display system where
the left- and right-eye images are alternated one pixel at a time to create a com-
bined image with a checkerboard pattern of left- and right-eye image pixels.
The checkerboard image is then presented using a digital micro-mirror display
(DMD) projection system and matching active eyewear.
3D A to Z 199
circle of isolation The area of sharp focus surrounding the subject of a ste-
reogram when set against a completely defocused background. Coined in
2010 by Indian stereographer Clyde DeSouza after circle of confusion, the con-
cept describing the area of sufficiently sharp focus that defines a lens’ depth
of field.
color rivalry; color mismatch When the colors of conjugate points in a ste-
reogram are contrary to what the brain expects given their parallax difference
or other available depth cues, leading to contradictory depth information in
the fused image.
conjugate points The pair of points in the left- and right-eye images that rep-
resent the same point in three-dimensional space. When each image is viewed
by the corresponding eye, the positional difference between the conjugate
points in the two images is interpreted by the brain as being caused by visual
parallax (disparity) and is converted into depth information. Also correspond-
ing points; homologous points.
D
depth budget 1. Disparity limit or fusible range; the total range of depth that
may be comfortably represented by a stereoscopic display system, from maxi-
mum negative parallax (distance in front of the screen) to maximum positive
parallax (distance behind the screen). 2. The cumulative amount of depth infor-
mation that may be depicted throughout a stereoscopic work. To avoid overex-
erting the audience, a limit may be placed on how much depth will be presented
in a work so that one scene with a great deal of depicted depth is balanced by
another scene elsewhere in the work with relatively little depicted depth.
depth change stress Discomfort caused by repeated changes in image depth
that are too abrupt, too great, or too fast. Generally caused by cutting between
shots with significantly different depth characteristics or by the intemperate use
of subtitles with variable z-depths.
depth cue—perceptual Information that helps the visual system determine
depth and distance.
depth discontinuity An abrupt and unnatural change in perceived depth
between neighboring or overlapping objects.
depth jump cut An abrupt, unexpected change in image depth between shots
in a stereoscopic work. The audience will require additional time to adjust to
the new shot. Also 3D jump cut.
depth map A graphical depiction of the distance between the observer (usually,
a camera) and each visible point in a scene. May be used when generating stereo
views for 2D to 3D conversion or autostereoscopic displays. In many cases, storage
space and transmission bandwidth can be saved by representing a stereogram as a
single-eye image and a depth map (2D depth) rather than two full-eye images.
depth nonlinearity A stereoscopic distortion where linear distances in the
recorded scene are not perceived as linear in a stereogram: physical depth is
recoded accurately at the screen plane (the point of zero parallax or image
convergence), but increasingly stretched towards the viewer and increasingly
compressed away from the viewer. This distorts motion as well as depth. For
example, an object moving towards the camera will seem to accelerate even
though it is moving at a constant speed.
3D A to Z 203
convergence when looking at a near object or when the left- and right-eye images
contain too much disparity.
disocclusion The recreation of hidden parts of an object based on visual
information in the neighboring areas. Often necessary in 2D to 3D conversion
or when selectively adjusting depth information in a stereogram.
disparity—binocular 1. The distance between conjugate points in a
left- and right-eye image pair that makes up a stereogram. Only dispar-
ity in the same plane as the viewer’s eyes (generally, horizontal dispar-
ity) contributes to depth perception. 2. The apparent distance between
an object and the plane of the screen, either closer to or farther from
the viewer. 3. A difference between left- and right-eye images that is not
caused by parallax. Such differences conflict with the depth cues in the
image pair. Minor disparities interfere with the 3D effect, while signifi-
cant ones can cause physical discomfort in the viewer.
disparity map A graphical depiction of the positional differences
between conjugate points in the two images of a stereogram. Used
to help identify camera misalignment, judge the relative distance
separating different objects, highlight areas of positive or negative parallax, etc.
display plane The set of conjugate points within a stereogram that have zero
parallax (they are in the same position in the left- and right-eye images) and
describe a two-dimensional plane that appears to lie upon the surface of the
screen. This plane lies between the audience space and the screen space. All 2D
images appear to lie upon the display plane. Stereograms may depict objects on
the screen surface (zero parallax), in the audience space in front of the screen
(negative parallax), or in the screen space behind the screen (positive parallax).
duration of comfort The amount of time during which one can view a ste-
reoscopic work without experiencing discomfort.
dZ/dt Parallax change; the amount of change in the parallax of conjugate
points in a stereogram over time, calculated by dividing the amount of change
in image depth (dZ) by the elapsed time (dt).
E
edge violation 1. When an object with negative parallax (seeming to extend
out of the screen) is cut off by the edge of the screen. The brain does not know
how to interpret this visual paradox. One expects far objects to be cut off by
the edge of a window, but objects closer than the window should not also be
cut off. This paradox leads to a visual/brain conflict and impairs the illusion of
depth in the image. 2. When an object lies within one of the monocular areas
along the edge of a stereoscopic view. (It will be depicted in one eye view but
not the other.) Also breaking the frame; window violation.
extrinsic camera parameters The parameters that define the position (loca-
tion and orientation) of a camera in relation to a world coordinate system: T,
3D A to Z 205
the translation (in x, y, and z space) necessary to align the world coordinate sys-
tem’s origin with the camera’s principal point (the point where the optical axis
of the lens intersects the image plane), and R, the rotation necessary to align the
x, y, and z axes of the world coordinates with those of the camera coordinates.
The extrinsic camera parameters change when the camera moves. Often used in
computer vision applications or when reconstructing three-dimensional objects
from a series of two-dimensional images. Also exterior orientation.
F
far point 1. The conjugate points in a stereogram that appear to be farthest
from the viewer. The far point can change from frame to frame, but should not
exceed the maximum allowable distance defined by the depth budget. 2. The
farthest distance that can be seen clearly.
field of depth; FOD The range of depth that a stereoscopic display can physi-
cally produce given a particular voxel size and the angular resolution of the
anticipated viewer.
field of view; FOV The largest solid angle that can fit within one’s vision or
can be captured by a camera lens, generally measured as a certain number of
degrees from side to side (less often from top to bottom). An important factor
in stereoscopic calculations.
G
geometric distortion Unequal magnification or reduction across an image,
including barrel, keystone, mustache, and pincushion distortion.
H
half-resolution;—3D;—encoding An image that only uses half of the avail-
able horizontal or vertical frame; stereoscopic encoding where the left- and
right-eye images are both included in a single frame with each eye image tak-
ing up only half of the available space. This allows both left- and right-eye
information to be encoded in a format that can be processed, recorded, and
transmitted using standard 2D systems and techniques, though a specialized
display is required for final stereogram presentation. Also split-resolution.
hyperfocal distance The length from the imaging plane to the nearest object
in acceptably sharp focus when the lens’ focal distance is set to infin-
ity. All objects at or beyond the hyperfocal distance will be in focus at
the same time. An important factor when establishing image depth of
field.
I
interaxial; IA Literally, between axes. A comparison of some characteristic
between two lenses, usually the interaxial distance in a stereoscopic camera sys-
tem (the distance separating the optical axes of the two lenses).
210 3D TV and 3D Cinema
K
keystone distortion An image malformation caused when the optical axis of
the camera or projector lens is not perpendicular to the surface of the subject
or screen, respectively. When these elements are out of alignment in a camera,
the portion of the image that is closest to the lens will be abnormally large
while the portion of the image farthest from the lens will be abnormally small,
causing the shape of the image to resemble the keystone of an arch. In a pro-
jector, the effect is reversed (close is small, far is large). May rarely be caused by
a misalignment between the imaging plane and the lens.
L
left eye The left half of a stereo image pair; the image intended to be viewed
by the left eye.
left eye dominance Using the left as the starting eye for a sequential 3D
application.
left field Information from a stereogram intended for the left eye only.
left homologue A point in a left-eye image that corresponds to a point in the
matching right-eye image.
212 3D TV and 3D Cinema
M
micropolarizer; micro-polarizer; μPol A very small polarizer, typically a circu-
lar polarizer one pixel tall running the width of a display. Polarized 3D displays
have an alternating array of orthogonal micropolarizers covering their surface,
generally in an alternating left circular/right circular pattern to match the polar-
ized glasses worn by the viewer. The two eye images are row-interleaved (spa-
tially multiplexed) and presented simultaneously with one eye image on the odd
rows and the other eye image on the even rows. This technique was originally
developed by Sadeg Faris, founder of Reveo, Inc., in 1994.
N
negative parallax When a point in the left-eye image is to the right of the
conjugate point in the right-eye image (and vice versa) in a stereogram.
Conjugate points with negative parallax will appear to extend out from the
screen, occupying the audience space. The viewer’s eyes must converge on con-
jugate points with negative parallax to fuse them into a single image.
nodal points The two points along the optical axis of a lens where the angle
of a ray of light upon entry is equal to the angle upon exit with respect to the
front and rear nodal points, respectively. The nodal points may not lie within
the physical lens, but if the front and back surfaces of the lens are in contact
with the same medium (usually air), then the nodal points will lie upon the
principal planes. The principal planes lie perpendicular to the optical axis in
a lens. When viewed from the front of a lens, an entering light ray will appear
to cross the front principal plane the same distance from the optical axis as it
appears to cross the rear principal plane. Depending on the configuration of a
compound lens, the principal planes may not lie within the physical lens.
nodal shift The changes to a varifocal lens’ nodal points as the focal length is
changed. A natural characteristic of physical lenses that must be simulated for
the virtual lenses used in computer graphics.
3D A to Z 215
O
occlusion A blockage or obstruction; when an object that is closer to the
observer blocks from view some portion of an object farther from the observer.
Commonly used as a monocular depth cue. Can cause a visual conflict if an
object that is supposed to be in the audience space is cut off by the edge of
the screen. (It seems as if the edge of the window surrounding the more dis-
tant screen has occluded the nearer object—a situation that cannot occur in the
natural world.) Also overlap.
occlusion information Stereographic metadata that describes the objects
within a scene, indicating whether specific points on each surface are visible or
occluded from each viewpoint from which images were recorded. These data
allow for adjustments to the binocular disparity for each depicted object, selec-
tively increasing or decreasing the apparent depth in a stereogram.
occlusion resolution The precision with which occlusion information is
recorded.
optic axis; optical axis 1. An imaginary line that extends through the
physical center of the lens, and on into the distance. This represents
the optical center of the image created by the lens. 2. The imaginary
line that passes through the center of the eye and represents the direc-
tion of one’s gaze. Also principle axis.
ortho stereo; orthostereoscopic Perfect stereoscopic reproduction; images
that appear as did the original recorded scene without any depth distortions.
With theoretically ideal stereoscopic viewing conditions, the interaxial dis-
tance used to record the stereogram matches the inter-ocular distance of the
observer; the images are presented using lenses of the same focal length as
those used when the images were recorded, and the stereogram is viewed from
a distance that ensures the image has the same visual angle for the observer as
was recorded by the camera.
P
page flipping Quickly switching between left- and right-eye images in a ste-
reoscopic display system. Viewed with active shutter glasses synchronized to
switch in time with the images.
Panum’s fusion area; Panum’s fusional area The region surrounding the
horopter where image disparity is small enough that the left- and right-eye
images can be fused into a single image. Beyond Panum’s fusion area the dispar-
ity is too great to fuse the two eye images, resulting in double vision, or diplopia.
First described by Danish physician Peter Panum.
Within the fovea (at the center of the retina), retinal disparity is limited to
about 0.1—if the conjugate points in the images recorded by the retinas are
any farther apart, they cannot be fused. Moving outward from the fovea by 6
increases the acceptable retinal disparity to about 0.33; at 12 from center,
216 3D TV and 3D Cinema
acceptable disparity increases to about 0.66. Extended viewing times and ver-
gence can extend the range of fusible disparities to 1.57 for uncrossed dispar-
ity and 4.93 for crossed disparity.
The physical position and size of the fusion area in three-dimensional space
depends on the vergence of the observer’s eyes. When the eyes are directed
at a near object, disparity increases greatly as an object moves away from the
horopter, narrowing the fusion area. When the eyes are directed at a distant
object, disparity increases slowly as an object moves away from the horopter,
extending the fusion area.
parallax The apparent displacement of a fixed object when seen from two
different positions or points of view at the same time; the distance between
conjugate points in the left- and right-eye images of a stereogram. Typically
measured in pixels (image resolution dependent), as a linear distance between
conjugate points (taking into account display size), or as an angle of so many
degrees (taking into account both display size and viewing distance).
Human binocular vision uses parallax to give depth perception. (Each eye
records a slightly different image of the same object because the eyes are sepa-
rated on either side of the nose. The brain interprets these differences to calcu-
late the distance to the object.) Non-reflex cameras suffer from parallax error
between the taking and viewing lenses. When the subject is at a significant
distance from the camera, this is not enough to notice. As the camera moves
closer to its subject, allowances may have to be made for the fact that the two
lenses see a slightly different image.
Stereoscopic imaging systems rely on the parallax between left and right eyes to
create their 3D images. In such systems, two images are separated by the aver-
age distance between a pair of adult human eyes—approximately 50–70 mm
(the exact distance is adjusted based on the composition of the subject being
filmed and its distance from the camera). When these two images are pre-
sented to the viewer, one to each eye, the brain interprets these slightly differ-
ent images to create the impression of three-dimensional depth.
In stereoscopic systems, parallax is characterized with respect to an object’s
apparent position with respect to the screen:
n positive parallax: The left-eye image is to the left of the right-eye image and
the object appears to be behind the screen (in the screen space).
3D A to Z 217
n negative parallax: The left-eye image is to the right of the right-eye image
and the object appears to be in front of the screen (in the audience space).
n zero parallax: The left-eye image is aligned with the right-eye image and the
object appears to be on the surface of the screen (equivalent to a traditional
2D presentation).
parallax budget The total range of parallax from largest negative (in front
of the screen) to largest positive (behind the screen) that may be comfortably
presented by a stereoscopic display system. Essentially the same as the depth
budget, but represented by amount of parallax rather than by apparent depth.
The total parallax budget is largely dependent on the average
viewing distance (increases with the size of the screen) and
the average viewer’s inter-ocular distance (larger for adults
than children).
parallel cameras A stereoscopic camera configuration
where the cameras are positioned so the lens axes are paral-
lel, resulting in a fixed interaxial distance to infinity. Parallel
images must either be cropped or recorded with imaging
grids that are off-center from the optical axis, since they do
not completely overlap. This avoids the geometric errors
common to toed-in cameras.
parallel rig A mechanical framework designed to support
two identical cameras set side by side. The optical axes
may be set parallel to one another or, if the rig supports
convergence adjustments, toed-in by equal amounts.
Percival’s zone;—of comfort The area within the visual
field where binocular images can be fused into a single
three-dimensional view without particular strain or dis-
comfort, defined as the middle third of the distance
between the nearest and farthest fusion points when the
eyes are focused at a particular distance. First proposed
218 3D TV and 3D Cinema
Q
quincunx 1. An X-shaped geometric pattern of five points—four at the corners
of a rectangle and the fifth at the rectangle’s center—common to the five-spot
on dice or dominoes. 2. An anti-aliasing pattern where the value of each pixel
is based on an average of five different samples taken in the shape of a quin-
cunx. The pixel itself is aligned with the center of the quincunx. The corner
samples are shared by adjacent pixels, so only twice as many samples as pixels
are required, rather than five times as many. 3. The pattern of left- and right-
eye pixels in a checkerboard stereogram, which can be taken to be a series of
interlocking X-shaped pixel groups.
R
reconverge To adjust the apparent depth in a stereoscopic image.
retinal rivalry The reception of images by the eyes that contain differences
between the left- and right-eye image that are not due to horizontal disparity
and interfere with stereoscopic fusion. This can lead to distraction, eyestrain,
headache, or nausea. Often caused by vertical parallax. Also caused by color
rivalry (different colors perceived by each eye) and luminance rivalry (different
levels of brightness perceived by each eye). Some rivalries occur naturally and
are not stereoscopic defects, such as iridescence, sparkle, and occlusion.
right eye The right half of a stereo image pair; the image intended to be
viewed by the right eye.
right eye dominance Using the right as the starting eye for a sequential 3D
application.
right field Information from a stereogram intended for the right eye only.
right homologue A point in a right-eye image that corresponds to a point in
the matching left-eye image.
row-interleaved;—3D A 3D raster image format where left- and right-eye
image information is presented in alternating rows within the image. By con-
vention, one starts with the left eye in the top row of the image and moves
down. May be used with an interlaced or progressive scan system. Generally
viewed on a polarized display. Also IL; interleaved; line-interlaced 3D.
220 3D TV and 3D Cinema
S
saccade The rapid eye movement necessary to take in a large or complex
image. The human eye only senses image detail at the center of the visual field.
Therefore, it is necessary to move the eye over a large object in order to take
in its entirety. On average, the eye moves once every 1⁄20 of a second—approxi-
mately the time span of persistence of vision. This is a subconscious action,
but it means that one cannot comprehend a large or complex image all at
once. These eye movements are not random and tend to follow visual patterns,
but different people will scan the same image differently, and therefore have a
different subjective experience. Sounds, unlike images, are taken in all at once.
There is no aural saccadic equivalent.
screen parallax 1. The physical distance between conjugate points when a ste-
reogram is presented on a particular size screen. 2. The plane of zero parallax
aligned with the surface of a display screen.
screen space The perceived area behind the plane of the screen in display/
viewer space. When objects in a stereogram appear behind the screen, they are
said to occupy the screen space.
screen width The horizontal dimension of a display screen. An important
factor in stereoscopic calculations.
segmentation The separation of overlapping objects at different depths in
a scene when extracting z-depth data or converting 2D to 3D. Outside edges,
where one object overlaps another, are most common, but inside edges, where
an object overlaps itself, can be particularly challenging. For example, when an
3D A to Z 221
actor’s arm extends towards the camera, it should be closer to the camera than
the torso, but it can be difficult to segment the arm from the torso when they
are both the color of the actor’s costume.
selection device The physical component that separates and directs the
left- and right-eye images of a stereogram to the corresponding eye. Auto-
stereoscopic devices include an integrated selection device (generally lenticular
or parallax barrier). Other technologies use some form of eyewear, either pas-
sive (anaglyph or polarized) or active (mechanical or liquid crystal shutters).
ChromaDepth and the Pulfrich effect use special types of selection device eye-
wear to create a 3D effect from a single flat image. Also analyzer.
separation 1. The degree of isolation between signals flowing in two paths; a
lack of crosstalk. 2. interaxial—The distance that separates the left and right
cameras in a stereoscopic process, measured along the optical axes of the cam-
eras’ lenses; interaxial distance. 3. The distance between the conjugate points in
a stereogram.
sequential stereogram A stereoscopic image where the left- and right-eye
images are recorded one at a time by the same camera, shifting the camera hori-
zontally by the desired interaxial distance between exposures. If a toed-in con-
figuration is desired, then the camera must also be rotated the correct amount
between exposures. This works best when the subject is not in motion (since
the two images are not recorded at the same instant in time) and when camera
movement can be carefully controlled.
Stop-motion stereography often employs a motion-control camera rig to
achieve precise horizontal separation of the left- and right-eye views, such as
for the movie Coraline (2009). For hand-held photography, the horizontal off-
set may be achieved by simply shifting one’s weight from one foot to the other
(the shifty or cha-cha method). This latter technique is not generally employed
by professional stereographers.
shading Slight variations in an object’s color or brightness that offer visual
clues to the object’s shape and size.
shear distortion A stereoscopic malforma-
tion that occurs when the viewer’s gaze is not
perpendicular to a stereoscopic display. As the
viewer moves progressively off-axis horizon-
tally, objects nearer than the screen (with nega-
tive parallax) will shift in the viewer’s direction
while objects farther than the screen (with posi-
tive parallax) will shift away from the viewer.
In addition to geometric shear, any movement
in the stereogram along the z-axis will take on
unintended movement along the x-axis.
After a short time, most viewers who are signifi-
cantly off-axis get used to the shear distortion.
222 3D TV and 3D Cinema
In presentations where the viewer is free to move about, shear distortion may
be used as a creative effect, causing still images to appear to move as the viewer
changes position. Anamorphosis can be used to correct for shear distortion by
recording the natural perspective of the depicted scene from the viewer’s off-
axis position. However, this only works from the one defined off-axis viewing
position and exhibits shear distortion from all others.
side-by-side; SBS;—3D A technique for recording stereograms where the left-
and right-eye images are compressed horizontally and presented next to each
other in the same frame.
For line-interlaced displays, every other line is discarded when the images
are expanded to their full width to fill one frame, resulting in individual eye
images that are ¼ the resolution of the original. Even so, in video applications
side-by-side 3D remains more popular than over/under 3D (which produces
½ resolution images), because side-by-side does not requires a frame buffer to
store the first eye image while the second is received, and it performs better
with MPEG encoding. Since side-by-side configurations preserve the vertical
resolution but reduce the horizontal resolution by half (and all stereoscopic
depth information is recorded in the horizontal direction), they generally do
not deliver the same depth representation as over/under configurations, which
preserve the horizontal resolution but reduce the vertical resolution by half.
side-by-side rig A mechanical framework designed to support two identi-
cal cameras set next to each other. The two cameras may be fixed in a sin-
gle position, or the rig may allow them to be moved horizontally to adjust
the interaxial distance or to be turned symmetrically inward from parallel
(toed-in) to converge at the desired plane of zero parallax. When toed-in, the
recorded images may contain geometric distortions that have to be corrected
in postproduction to ensure proper registration when presented as part of a
stereogram.
stereo extinction ratio The degree to which the left- and right-eye images in
a stereoscopic display can be seen by the opposite eye; a measure of crosstalk,
perceived as ghosting.
Stereo Image Processor; SIP The trade name for an electronic device pro-
duced by 3ality Digital, designed to perform a number of stereogram adjust-
ments in real time, including image synchronization, geometric and color
correction, image alignment, and output to anaglyph, frame-compatible, and
frame-sequential display formats.
stereo2Z The images and depth data obtained from two separate cameras
recording the same scene.
stereodepth; stereoZ A left- and right-eye image pair with depth data
that describe the three-dimensional nature of the objects depicted.
bounded by the flat-sided pyramid that extends from the center of the viewer’s
gaze through the edges of the screen.
T
temporal multiplexing; temporal separation Separating the individual views
in a stereogram or multiview 3D image in time and presenting them in sequence.
time-sequential;—3D A stereoscopic display system where left- and right-eye
images are presented one at a time in a high-speed alternating pattern. Only
one eye is receiving an image at any given time, but the switching is so fast that
this is imperceptible to the viewer. Active or passive eyewear ensures that each
eye sees only the correct images. Polarized time-sequential systems alternate
the polarization direction as each image is presented, while active systems turn
the lenses in the eyewear on and off in time with the left/right image sequence.
3D A to Z 227
By convention, one presents the left-eye image first, followed by the right-eye
image.
n For interlaced video display systems, the eye images are in alternate fields of
the same frame (field-sequential).
n For checkerboard systems, the eye images are combined into a single frame,
alternating the presented eye with each pixel.
n For all other time-sequential display systems, the eye images alternate with
each frame (frame-sequential).
toe-in The degree to which the lens axes of a pair of stereoscopic cameras
are turned inwards from parallel so that they converge at a distant point. Also
angulation; stereoangle.
Toed-in cameras are most often used for natural imaging (photography with
a physical camera) when the cameras are too large to fit next to each other at
the desired interaxial distance, but this introduces geometric errors that reduce
image quality and must be corrected in postproduction.
U
unfusable images Left- and right-eye images in a stereoscopic pair that con-
tain conjugate points with excessive parallax, either positive or negative, such
that the individual images cannot be fused into a 3D view.
V
vergence;—eye movement The coordinated inward (convergence) or out-
ward (divergence) turning of the eyes when focusing on an object or when
adjusting one’s gaze from one point to another point at a different distance.
This has the effect of projecting the object of interest onto the fovea at the cen-
ter of the retina where clear, full-color vision is possible. Vergence is only effec-
tive for objects up to about 200 meters away (stereo infinity). When viewing
objects beyond this point, the eyes are held parallel. Unlike other types of eye
movements, with vergence, the eyes move in opposite directions. Also angula-
tion; stereoangle.
vergence point The point in space where the lens axes of eyes or toed-in
cameras cross. This also defines the zero disparity plane for images recorded
with toed-in cameras.
vertical alignment error; vertical misalignment A vertical deviation between
left- and right-eye images in a stereogram where one of the images is higher
than it should be so that horizontal points do not align.
vertical disparity; vertical error; vertical offset; vertical parallax The verti-
cal distance between conjugate points in the left- and right-eye image pair that
makes up a stereogram. Only disparity in the same plane as the viewer’s eyes
(generally assumed to be horizontal) contributes to depth perception, so verti-
cal disparity should be avoided. Small amounts can be ignored, but too much
vertical disparity will interfere with three-dimensional image fusion, leading to
viewer discomfort and double vision. Also height error.
3D A to Z 229
view frustum; viewing frustum The flat-top pyramid that encompasses the
volume of space recorded by a camera, particularly a synthetic camera record-
ing a computer-generated image.
The view frustum through a lens is conical, but photographic images are
recorded with a rectangular aspect ratio, so a camera’s view frustum is shaped
like a pyramid. The front plane of the view frustum is
defined by the closest object visible to the camera. In
computer graphics, near objects are excluded from view
by a clipping plane so the computer does not have to
perform unnecessary calculations. Similarly, the back
plane of the view frustum is defined by the farthest
object visible to the camera or the back clipping plane
in a computer-generated image.
viewing angle 1. The figure formed by extending a line from the center of
the viewer’s gaze to the surface of the screen, measured in relation to both its
horizontal and vertical offsets from a line perpendicular to the surface of the
screen. 2. The horizontal angle of view when observing a display screen. An
important factor in stereoscopic calculations. Calculated as a function of the
screen width and viewing distance and measured in degrees.
230 3D TV and 3D Cinema
viewing distance The length from the observer to the display surface. An
important factor in stereoscopic calculations. Generally measured from the
center of the viewer’s gaze to the center of the screen.
viewing zone The volume of space within which a viewer’s head must be
placed in order to perceive the full stereoscopic effect. A particular display tech-
nology may have one or more viewing zones, often a central zone and side
zones. Also head box; sweet spot.
viewing zone angle A measure of the extent of the viewing zone measured as
an angle from the center of the display, either horizontally or vertically.
Z
z-depth The horizontal offset between conjugate points in the left and right
images in a stereogram. Larger z-depth values result in an increased 3D effect.
Positive values provide depth away from the viewer while negative values pro-
vide depth towards the viewer. Also 3D depth.
zero binocular retinal disparity When the images recorded by the left and
right retinas in an animal with binocular vision are perfectly aligned, resulting
in zero parallax between the conjugate points.
zero disparity plane; ZDP The distance in a scene at which objects will
appear to be on the screen in the final stereogram. The ZDP divides a scene
into objects that will appear in audience space (in front of the display surface)
and in screen space (behind the display surface). Also HIT point; plane of con-
vergence; zero parallax depth.
zero interaxial When the two images recorded by a 3D camera system exactly
overlap—there is no interaxial distance separating them. The greater the sepa-
ration between the images, the more pronounced the 3D effect. At zero inter-
axial, there is no 3D effect.
232 3D TV and 3D Cinema
zero parallax When a point in the left-eye image is perfectly aligned with the
conjugate point in the right-eye image in a stereogram. Conjugate points with
zero parallax have no stereoscopic depth information and appear to be at the
same distance from the observer as the screen.
zero parallax setting; ZPS Adjusting the relative horizontal offset of the left-
and right-eye images in a stereogram so that the conjugate points that lie on
the intended display plane have zero parallax.
RICHARD W. KROON
Richard W. Kroon is the Director of 3D Services for Technicolor’s Media Services
Division in Burbank, California. In addition to being an award-winning videographer
and published author, Mr. Kroon is a member of the Society of Motion Picture and
Television Engineers (SMPTE), National Stereoscopic Association (NSA), International
Stereoscopic Union (ISU), and Dictionary Society of North America. He is a Certified
Project Management Professional (PMP) and PRINCE2 Registered Practitioner.
Mr. Kroon holds a BS from the University of Southern California (USC), an MBA
from Auburn University, and post-graduate certificates in Film, Television, Video, and
Multimedia from UCLA Extension’s Entertainment Studies program.
Richard Kroon
Index
233
3D-compatible projectors, 73t overview, 181, 184–189 correction of depth accident on,
3D-matched zooms, 18 replicating 3D world or human 87
3D-one camcorder, 52 vision, 188–189 directing live 3D sports, 101–102
3G HD-SDI, 98, 98 screens, 185 Gerhard Lang interview, 108–113
4% depth budget, 170 screens versus windows, 187 graphics, 106, 107–108
4D, defined, 196 soft or strong stereoscopic effects, overview, 10–12
10 rules of 3D, 17 187–188 recording 3D, 113–117
50% mix monitoring, 60, 61f stereo stress as storytelling tool, examples of 3D recorders, 115
2011 Consumers Electronics Show, 186–187 recorder functions, 113–114
Las Vegas, 7 stereoscopic window, 185 recording metadata, 114–115
assistant stereographer, 159 safe volumes for graphics, 107
A audience space, 185, 197, 197f active occlusions processing,
accommodation, defined, 196 Austin, Bruce, 19–20, 114 107
accommodation/convergence reflex, automation, 168 passive volume segmentation,
196 Avatar, 2–3, 8–9, 186, 187, 107
accommodation/convergence 188, 189 setting up camera unit, 92–97
relationship, 196 beam-splitter configurations,
accommodation/convergence rivalry, B 92–93
196 Bailblé, Claude, 149 calibrating cameras, 96–97
accommodation/vergence reflex, 196 banding, 218 color-coding, 93
accommodation/vergence barrel distortion, 197 registering cameras. See
relationship, 196 beam splitter registering cameras
accommodation/vergence rivalry, common characteristics of, 39–40 setting depth, 97
196 common problems with, 39–40 sports
accuracy configurations, 92–93 comparison of 2D and 3D
extending high-end rigs, 87 defined, 197, 197f direction, 162t
importance of, 15 testing, 40 effect of 3D on, 20
active 3D TV, 73t Bergeron, Michael setup for 3D, 15–16
active depth cuts, 104 maximum sustainable retinal tools for, 102–106
active occlusions processing, 107 rivalry, 15 3D outside broadcast. See
active stereo, 6–7, 196 Panasonic AG-3DA1, 50–51 outside broadcasts
ActivEyes, Volfoni, 7 Berman, Art, 4 depth continuity, 104–105
adjustable mechanical controls, 32 Binocle Disparity Tagger, 79, 80f, switchers. See switchers
AG-3DA1, Panasonic, 50, 138–139 130–131, 131f transporting to vision mixer or
AG-HMX100 switcher, Panasonic, Binocle geometric corrections, 89 recorder, 97–101
102 binocular, 197 frame-compatible formats. See
Alexa, ARRI, 46 binocular depth cue, 197 Frame-Compatible Format
Alonso, Jordi, 133–136, 182 binocular depth perception, 197 (FCF)
alternative 3D content, 12 binocular disparity, 197, 197f stereoscopic HD-SDI, 98–99
anaglyphs, 196 binocular rivalry, 198 BT-3DL2550 monitor, Panasonic,
analyzer, 221 binocular suppression, 198 74, 74f
angular disparities, 21 binocular vision, 198 by-appointment entertainment, 8–10
angulation, 227 binocular vision disability, 198
aperture, 55–57, 81 bit stream conversions, 75 C
Apple Final Cut Pro, 118 Black Eyed Peas, 10, 11f cable, 3D distribution by, 126
Arnheim, Rudolph, 141–142 Blu-ray 3D, 126–127 cables, 98, 98
ARRI Alexa, 46 brain shear, 198 calibrating cameras, 96–97, 111
Arri Depth Chart, 165f breaking the frame, 204 CamAlign 3-D Pilot, DSC Labs, 94f
artistic challenges, 183–189 breakout circuit on NOC, 83 camcorders, 3D, 49–52
audience space, 185 brightness, 77 3D-one and Frontniche, 52
depth continuity, 185–186 broadcast cameras, 45 Panasonic AG-3DA1, 50
depth placement of action and broadcasts, 93 Sony, 51–52
characters, 188 comfortable 3D placement in camera coordinate system, 198, 198f
geometrically realistic depth, 184 coherent space with live camera parameters
nonrealistic depth, 184 picture, 106–107 extrinsic, 204–205
Index 235
color balancing, 169 single-lens depth acquisition stereopsis, 15, 225, 225f
fast tracking camera setup, 168 techniques, 57–58 stereoscopic, defined, 225
image processing, 167 single-lens parallax generation, stereoscopic art development, 8–10
need for multiple cameras, 169 55–57 stereoscopic calculator, 225–226,
SIP, 167–168 single-projector system, 223 226f
zoom lenses, 168–169 SIP (Stereo Image Processor), 79f, stereoscopic graphic system, 109f
overview, 163, 163f 80f, 167–168, 224 stereoscopic multiplexing, 59, 226
use of long lenses, 21 slide bar, 223 stereoscopic rendition, 226
scientific challenges, 189–194 slow motion, 103 stereoscopic supervisor, 25
3D fatigue, 190–192 SMPTE 292M, 98 stereoscopic window, 185
duration effect, 191–192 SMPTE 3D Home Master, 124 SterGen real-time converter, 104
the focus/accommodation SMPTE 372M, 98 storage disks, RAID-arrayed, 115
reflex, 190–191 SMPTE 424M, 98 storytelling
vertical parallaxes and other soft stereoscopic effects, 187–188 impeding, 19
factors, 191 software 3D recorders, 116 mixing wow-ing shots and, 176
overview, 181–182 solid state disk (SSD) recorders, 115 not dependent on 3D, 21
screen size and human brain, Sony stereo stress as tool, 186–187
192–194 camcorders, 51–52 strong stereoscopic effects,
experience changes with screen geometric corrections, 89–90 187–188
size, 193–194 Sony HDCP1, 47–48 surface mirror, 38, 83
screen size and maximum Sony LMD2451TD monitor, 74 sweet spot, 230
parallax, 193 Sony Vegas Pro, 118 switchers, 102
screen parallax, 220, 220f, 222 spatial capture, 223 3D conversion, 103–104
screen size spatial phase imaging (SPI) depth of known geometries, 103–104
and human brain, 192–194 acquisition, 58 of large landscapes, 103
experience changes with screen special effects (SFX) supervisor, 158 rules for, 104
size, 193–194 split-resolution, 208, 208f stereoscopic replays and slow
screen size and maximum Spottiswoode, Raymond, 142 motion, 103
parallax, 193 SSD (solid state disk) recorders, 115 synchronization error, 226
importance of, 21 Stassen, Ben, 187–188
screen space, 220 static depth cue, 223 T
screen width, 220 static interposition, 223 TAB (top-and-bottom) frame-
screens versus windows, 185, 187 STB (set-top boxes), 4, 126 compatible formatting, 99,
Sean Fairburn registration chart, 93 steady-cam rigs, 41–42 100, 100f
seat kills, minimizing, 92 stereo, 224 tags, disparity, 78–79
segmentation, 220–221 stereo acuity, 224 tape recorder, 116–117
selection device, 221 stereo blind, 6, 224 TC (time code) sync, 125
Sensio FCF format, 100–101 stereo extinction ratio, 224 technical challenges, 182–183
separation, 221 Stereo Image Processor (SIP), 79f, impact of technological
sequential stereogram, 221, 221 80f, 167–168, 224 constraint on creativity,
set-top boxes (STB), 4, 126 stereo imaging device, 224 182–183
setup of equipment, 15–16, 168 stereo infinity, 224 overview, 180–181
SFX (special effects) supervisor, 158 stereo stress as storytelling tool, tools for creativity, 183
shading, 221 186–187 technician stereographer, 158
shear distortion, 221–222, 221f Stereo Tango alignment procedure, Technicolor film-based 3D
SI-2K, Silicon Imaging, 46 93–96 projection system, 128
SI-3D, Silicon Imaging, 46 stereo window, 224–225 Technicolor Stereoscopic Certifi3D,
side-by-side (SBS), 99, 100, 100f, stereo2Z, defined, 224 63f
105, 222 stereodepth, defined, 224 Telecast fiber converter, 98–99
side-by-side rig, 37, 37f, 222–223, Stereo3D Toolbox, 118f television content, defined, 2
222f stereoangle, 227 temporal mismatch, 226
Silicon Imaging SI-2K, 27f, 46 stereogram, sequential, 221, 221 temporal multiplexing, 226
Silicon Imaging SI-3D, 46 stereographer, 23–24, 171, 225 testing beam splitter, 40
single-lens camera, 223, 223f stereography, defined, 225 theater, defined, 2
242 Index