Stereo convergence

[GCharb]

Hello all

Still working on Stereo rendering here, like I said in a previous feature request, 3D rendered movies are becoming a must in todays 3DTV world.

Predicted 4.2 millions of 3DTV sales in 2010 and 12.9 millions in 2011, makes 3D rendering a tool that has to be available to any animator.

As it is, AS Pro has basic anaglyph rendering, which is nice but it needs better tools, like rendering of seperate images but also a way to control stereo convergence, or zero parallax as they call it.

What this means is this, you render your scene in a way that the center point or the focus of the scene is 2d, the front sort of jump out of tv and the back goes farther in the back.



The green line is the zero parallax, where all is 2d, or flat on the screen.

All that is closer to the camera seems to pop out of the screen while all that is further then the zero parallax point seems further in the screen.

G

[Rhoel]

Ah, the convergence debate

This is one of the most debated subjects in 3D. To converge or not converge. Its actually a remnant of the old film days when very limited post production was available. Today, we have a raft of tools to play in the digital back office.

The two strongest arguments against convergence is the danger of infinity divergence, and of image keystoning. Both mess with the brain and cause headaches.

But with parallax, you can never have infinity divergence or keystoning.

However, you do need to overshoot the image width to allow for post production depth resetting (float etc). This is achieved by sliding the left and right images towards each other - this is a vital stage for matching the scene to scene cuts. As in CinemaScope where its really bad to have subjects switching from extreme left to right shot-to-shot, cutting from different scene depths is not good for the audience. So the idea is to make planes of interest the same.

In this example, the left image is the original, with guy set back from the screen space. The right is the adjustment. The black lines are the camera axis in parallax, the red and green lines are the unchangeable angles to the object IN THE RECORDED IMAGE (ie, not at the point of shooting - this sis post-production).



The right image will be a better cut as he is now in the same plane of interest as the previous shot: The cup is now floating forward of the screen-space, drawing the audience into the shot.

This middle-adjustment is actually very easy but is seriously difficult for noobs to get their head around as there is no equivalent in the human brain - the eyes are fixed distance, and there is no intermediate image process before the data reaches the visual cortex. But then, the eyes can't zoom either but the brain has no problem when viewing such tweaking optical data.

Download Stereo Photo Maker, load a stereo image and hit F5 - then use the left and right arrow keys to move the point of interest into a "mono" position: That's the screen space.

If you go see Avatar 3D, take your glasses off and the left/right merged images unprocessed, you'll see this is what Cameron is doing - no major ghosting on the character.

Also, this shooting over-width is the same techniques used in the FujiFilm W1 digital 3D stills camera. They align the image in memory then throw away and edge safety they don't need when they save.

Hopefully that's as clear as mud as usual - I like consistency

Rhoel

[Rhoel]

Keystoning is a problem which occurs when a camera is not directly over the vertical access: One side to the object is closer than the other.



In the example, we have a rectangle. The left camera is colour coded pink, the right camera image green. When the two are overlaid, it is easy to see the mismatch. This can be correctable in post - afterall, its a prblem with live action shooting..

But in animation, when there are alternatives, why introduce an error which doesn't have to be there?

Rhoel

[Rhoel]

To try and show how depth correction works (without the use of 3D glasses), I have output an image of a jet fighter, setting the point of interest at three different points. I am using wiggle3D to show the effect. The Image rotates about the point of interest

- please remember, the camera original is the same for every image - this is not in-camera but a post-production adjustment.


The first is set on the nose cone of the Jet.


This is set on the cockpit, the most logical point.


This time, the POI is set on the tail/rear of the aircraft.

As you can see from the three images, one original, three different ways of balancing the shot. All are the same image size.

You could not want to cut from the first to the last image then back, the audience would be yo-yoing their eye focus back and forth.

It would be tiring if that happened over a 90 minute film.

Rhoel.

[GCharb]

Thanks for the extra infos!

G