3D IMAX camera

[zak forrest]

---

[Joseph Ward]

Why don't they just use Epic 617?:001_cool:

[Samuel Martin]

Has RED fully committed to support 3D acquisition on the 617? Because unless they do so...

[Iannis Holwech]

Why don't they just use Epic 617?:001_cool:

I guess you mean the 645. The 617 is much too wide for IMAX 1:44:1 AR.

[Austin Glass]

I guess you mean the 645. The 617 is much too wide for IMAX 1:44:1 AR.


No he means the 617. He's suggesting recording two images side-by-side onto the same sensor. It's really quite brilliant as long as there are no over-heating issues.

[Samuel Martin]

That's what I mean, there is a lot of talk and a lot of people claiming it is a great idea but unless it is developed by the manufacturer, in this case RED, there is no way you will ever see it happening even if anyone comes up with a way of doing it.... Not on commercial work... And then, even if they do you might still need to use a side by side rig or mirror rig depending on how close or far apart the each eye's image needs to be.... Again, the one camera do it all doesn't exist...

[Jeff Kilgroe]

Using the 617 for dual-lens 3D has been discussed here ever since that brain was first announced. It makes a lot of sense, but we also have to consider how everything keeps changing and how the Monstro sensors and those other brains are probably still a ways off in the future.

Giving the impending surge of 3D production, I'd like to see a dedicated 3D brain from RED that makes use of two 2/3" sensors with dual lens mounts.

In the near future, I think the fixed-lens Scarlet is going to be a runaway success for 3D. Easy to link and control two of them and they have complete fly by wire controls for that lens.

[Jeff Coatney]

IMAX 3D still requires the 6.5cm inter-occular distance. The beam splitter rig is the best way to achieve this using large format imagers. Putting 2 lenses side by side on the 617 will likely induce headache and eyestrain. The central axis of each lens must be exactly 65mm apart at the nodal point for adjusting the convergence angle. This can be done using a prism, with the lenses facing each other, or with the lenses at 90 degrees through a beam splitter rig. This inter-occular distance and the much preferred deep focus makes the 2/3" sensor the ideal imager to replicate human vision, which is the real goal of 3D stereography. You don't shoot 3D like you would a normal movie. Also, single lens systems can't replicate this inter-occular distance organically, so the 3D effect of such systems is diminished considerably-- its fake 3D. You must have two lenses, each 6.5cm offset left to right with at least 10 degrees of adjustable convergence. Preferably with two cameras.

[Samuel Martin]

I couldn't explain it any better myself, Jeff... hats off!

[Jeff Coatney]

Thanks, Samuel.

[Ace]

with at least 10 degrees of adjustable convergence. Preferably with two cameras.

Swivelling sensor blocks (attached to the mount) could solve this on a single dual-lens/sensor camera... I wonder how that panasonic 3d camera handles the convergence thingy.

[Gavin Greenwalt]

Is optical convergence actually important? Wouldn't an X-Axis image shift accomplish the same thing in post?

[Jeff Coatney]

Is optical convergence actually important? Wouldn't an X-Axis image shift accomplish the same thing in post?

Optical convergence is the optimal method to create depth. Its the convergence angle in conjunction with the inter-occular distance that creates the 3D effect. X-Axis shift creates an illusion of 3D, but it can also induce eyestrain and headaches, because you're essentially mushing your two eyes together. While you can converge the subject with an X-axis shift, the space in front of the subject and behind the subject becomes a view that cannot be replicated in nature and thus can play havoc with your brain's desire to reconcile the image.

Real, organic convergence angle (physically separated nodal axes) is critical in the "Intimate Zone" of 0" - 24" from the camera and in the "Interpersonal Zone" from 2 feet to 10 feet from camera. These zones are where the 3D effect is greatest, and beyond this distance the stereo effect diminishes exponentially. Effective management of the convergence angle is most important in the intimate and interpersonal zone to avoid eyestrain and headache. The best rule of thumb is to ask yourself if the human eye can actually physically perform the movement you want the lenses to make.

I'd like to point out that 3D as we know it replicates human vision. If we were to put glasses on an animal like a horse or an elephant, the 3D effect would create massive disruption of their brain's ability to reconcile the stereo pairs, due to the differences in their inter-occular distance. This is why the inter-occular distance is 65mm, because that's the average distance separating the eyes of human beings. Supposedly, not everyone will have eyes at exactly this distance apart and those people will not be able to physically cope with 3D regardless.

[Gavin Greenwalt]

Thanks! I was mucking about with a little 3D side project on Friday and I wasn't sure whether it would make a difference between changing the angles of the cameras and just shifting one of the stereo pairs and cropping the result.

I couldn't tell a difference but I was just using Red/Cyan glasses so eye strain was inherent to the system. :D

[Jeff Coatney]

Yes, the anaglyph system can give you 3D, but the optical processing centers in the brain want to resolve the full spectrum of visible light per eye. When those RGB values don't correspond from what the left eye is seeing versus the right eye, the theory is that the brain does its own error correction and uses the values its given to resolve an image. Its not surprising that very real physiological effects can be experienced if the viewer is left in an anaglyph viewing state for long periods of time. Avatar would not be able to sustain its 2 hour 40 minute running time if it used the anaglyph system, the audience would likely have to walk out.

[Michael Hastings]

...You must have two lenses, each 6.5cm offset left to right with at least 10 degrees of adjustable convergence. Preferably with two cameras.

Jeff:

Do the beamsplitter rigs from ParadiseFX, Element, P&S, etc. do convergence? (and I'm understanding convergence as toeing in i.e. instead of the cameras being parallel they toe in slightly to converge on the subject distance.)

Because I have been told by 3D "gurus" to not worry about convergence - that intraocular distances are very important but just to keep the cameras parallel and handle convergence issues in software (because toeing them in creates more problems (with foreground/background?) than it solves.

[Jeff Coatney]

Jeff:

Do the beamsplitter rigs from ParadiseFX, Element, P&S, etc. do convergence? (and I'm understanding convergence as toeing in i.e. instead of the cameras being parallel they toe in slightly to converge on the subject distance.)

Because I have been told by 3D "gurus" to not worry about convergence - that intraocular distances are very important but just to keep the cameras parallel and handle convergence issues in software (because toeing them in creates more problems (with foreground/background?) than it solves.

I'm not intimately familiar with those rigs, so I can't speak to their capabilities, but as far as I am aware they do some minor degree of convergence. The "toeing in" (good descriptor) is critical to effective 3D, but the envelope of movement here is very limited. Most 3D experts tell you to avoid it because it is too easy to go past the physiological limit that the viewer can tolerate. If you used no convergence at all, you would only have the 3D effect within a fixed plane in front of the camera and you would have to keep your subject within those limits. This is do-able, but limits your blocking of actors and makes camera movements on the Z axis limited.

[Jeff Coatney]

Managing convergence on set will be easier when it becomes automated. I think we're due for some wireless follow focus systems that can be programmed to extrapolate the correct convergence angle derived from the lens data. These are proprietary and custom right now, but they'll soon leak out into the world, then it'll be fire-and-forget.

Basically, they take the distance calibration and piggy-back the values of the subject's distance from the lens and correct the convergence angle on the fly. The convergence angle can be easily extrapolated from distance the lenses are focused, because two of the three values are fixed-- its simple triangulation.