For the Parallel camera zealots.

[Okta]

http://video.google.com/videoplay?docid ... 3911842846#" onclick="window.open(this.href);return false;

Notice the cameras converging? Amazing eh? Worked great for Avatar, why not for the rest of us?

[Freke1]

It may indeed be time for a converging cameras experiment.
His cameras also move apart when the object comes closer I noticed.
I would imaging the background will be unwatchable with converging cameras so a shallow depth of field is required. And my tiny digital cameras can't do that.

[Likay]

Well, Simply: It's not easy trying to manage convergence while trying to take shots in real life. So keeping cameras parallell all the time simplifies things. It's also very easy to batchconvert without having to edit and align all pictures later on. The later adjustment of the pictures is actually more important than trying to maintain a proper convergence when shooting.

Note: When keeping the cameras straight parallell the resulting converted image will always appear at prescreendepth. Far away object are very close to screendepth. Of course in the readymade image it's desired that far away objects are further away inside the screen and therefore adjustments are necessary for each and every image.

Just a theory i have yet to try: If a convergence setting should be used for cameras:
Take a stereopicture on a very far away object with some amount of convergence on the stereocamera.
Assemble the images in the computer without horisontal realignment and check if the far away object is rendered at a proper backdistance (for 22" screen this should mean at most 2 centimetres imageshift but keep it less!).
Example: if the distance is shifted 2 centimetres on a 22" monitor the same shift will be 20 centimetres on a 220" screen! In other words impossible to pleasantly view without horisontal correction on the big screen.
If you have too little shift then increase camerasconvergence more and go for another shot until you get right.

By using some amount of convergence you'll save and increase some amount of pixels which go lost when aligning the images later on. I don't think anything will change in the perception of stereo by using convergence since one image simply is shifted sideways by rotating the cameras.

[cybereality]

Bottomline is our real eyes do not see in parallel. We must focus (converge) on an object/person in our view. That is how the human visual system works, I see no reason why this wouldn't be the way to shoot. That is how we see real life. That is how we game. And now, after seeing Avatar, that is how we watch movies. Why are people still on the parallel boat?

FYI: All my YouTube videos are shot with a slight toe-in and I think it comes out better.

[Likay]

True. But using too high convergence disables you from seeing further away objects if it's set too high (eyes have to converge outwards. ). Especially when trying to watch on larger screens.
In reality the brain chooses what to see and look at and it would be a pity only being able to watch only a part of the photo while everything in the background cannot be focused on.
In avatar cameron sometimes decided to blur the background to put even more focus on what he wanted us to see. It works very well in this matter though.
I'll try to find the angle for the cameras when an image still is pleasant to watch on the big screen. Settling for this will give problems on a bigger screen but at least it should give a hint about which maximum degrees to have the toe-in. It's no worse than you always can change the horisontal alignment later on in a viewer.
Using toe-in saves some part of the image since images taken in parallell always have to be adjusted to show depth and therefore looses small parts at the sides. It's a little unecessary.
Hypothetically: With an infinite viewing distance and a infinitely big screen the cameras cannot have any other position than straight parallell which is probably why cameras should be aligned in parallell. However, screens are of course limited in size...