Dave Forsey and Flip Philips have been discussing whether you converge
the two cameras in 3D, or whether you leave their axes parallel.
At least in the case of Transitions, the Expo '86 IMAX 3-D movie, sometimes
they did and sometimes they didn't. There are tradeoffs.
Your eyes are used to converging and focusing at the same time. You can
maintain this link in 3D only if the object you are looking at appears
at the same distance as the screen, with the rest of the scene extending
in front of and behind the screen - pretty limited. However, your visual
system is pretty flexible, within limits, and you can actually present
scenes where the eyes focus separately from their convergence, provided
you don't try to have them converge *too* far, or try to make them
*diverge* rather than converge.
Good 3-D gives you the feeling of "being there", of seeing a real
three-dimensional environment. For this to work, you eyes must be
able to roam around the environment without finding things that just
"don't work" to your visual system. This also means that large depths
of field are necessary, requiring small lens apertures on the cameras
and lots of light.
In the real world, as your eyes roam around a scene, they change convergence
as they change focus, and the object being viewed is always at the point
where the eyes converge. The cameras cannot do this, so convergence has
to be set to suit the subject matter of the scene.
In scenes that contain far-away objects, you almost always want the
cameras parallel, since that gives superimposed images on screen for the
far-distant objects. The audience can thus look at the far objects
naturally, without strain. If the cameras were converged, the eyes of
the audience would have to *diverge* to look at the far-away objects, which
most of us are not capable of.
However, really close objects then have a large offset between the two
images on screen, and it becomes impossible for some members of the
audience to converge their eyes enough to see both images simultaneously.
This effect is worst for people at the front of the theatre, since a given
linear displacement in the images translates to a larger angular displacement
So, in scenes where the focus of interest is quite close to the cameras,
the cameras are converged somewhat to reduce the on-screen displacement
and help out the audience. This works fine if the scene is such that
there isn't any detail at infinity, or if you know the audience attention
is going to be riveted on the thing in the foreground (e.g. the aforementioned
egg hovering above the head of the person in front of you in the audience).
It's all illusion anyway - you can't reproduce mathematically correct
3D for more than one person in the audience anyway, so what you try for
is the best apparent 3-D for the most people simultaneously.
For example, when the egg breaks in Transitions, some people think it's
going to land in their lap, and some just watch it land on the head
of the person ahead of them without worrying about their own clothes.
Another example of this: the 3-D effect varies greatly with distance
from the projection screen. A given angular displacement of objects
results in an apparent depth which is a particular fraction of the distance
between the viewer's eye and the screen. This is true for differential
depth as well - an object which appears to be 1 meter deep to someone
at the front of the audience will appear 2 meters deep to someone at the
back, if they are twice as far from the screen. Now, when you consider
that apparent height and width of objects is inversely proportional
to distance from screen but apparent depth is directly proportional to
distance, then you can see that the apparent depth-to-width ratio
of objects changes as the square of the viewing distance change.
Often, you brain compensates for this, since you *know* the proportions
of things, but sometimes it is very obvious. In Transitions, there is
a scene where a small boy is asleep in bed. From the front of the
theatre, the bed looks rather short compared to its width, but well it's
a child's bed so that's OK. But from the back of the theatre, it looks
like the bed is about 15 feet long.
In practice, the 3D is usually computed so that things look "right" for
someone in the centre of the audience, and so people at the front see
compressed depth and people at the back see it exaggerated. (This assumes,
of course, that the cinematographer or "3D consultant" understands the
mathematics of 3D at all - if they don't, who knows what you get).
I watched "Transitions" many times with the theatre empty, and found that
although the 3D looked more natural from the centre, I preferred either
the extreme front or extreme back of the audience seating. From the
back, you get the most exaggerated 3D, and from the front you get the
most effect from the IMAX screen filling your field of view, towering
above you. I remember one scene where two boys were shovelling wheat
in a truck, and I was afraid that they were going to bury me in it at