Stereoscopic Camera Rigs

[Varinder]

I just found this plugin for LW for stereoscopic work while searching around the cgtalk forums.

http://colm.jp/plug/stereo.html

It looks like an interesting plug-in and I'm testing it out right now.

[Varinder]

I just tried out the plugin and it works pretty well. Since it renders both of the left and right camera angles at once which is pretty convenient. Although you would have to render out the width of your shot by double in order to get the frame size you want.

e.g. 1280x720 would have to be rendered at 2560x720 in order to get both left/right camera angles in the same frame size when there combined.

There was another tool/plugin i forgot to mention that would be nice to have for stereoscopic work. At the moment you can only look through one camera angle at a time. But it might be pretty handy to be able to see both left and right camera angles at once in different viewports to see the difference between both cameras. This may not be a very important tool/plug-in but could be handy at times.

[Eugeny]

As far i know this pludin (as well as LW native stereoscopic render) use parallel cameras to achieve stereo effect. Some times it's useful but some times u have to deal with strong differences on the left and right edges of left and right frame. For this case using two cameras with one target can help.
It's very simple to set up stereoscopic cameras rig:
Parent two cameras (left and right) to one (it's will be central) camera
Make simple expression that copy negative position X of Right camera
Apply that expression to Left camera, now u can move only Right camera the left one will move automatic to opposite direction.
Make null and and make it child of Central camera, move it forward in Z direction
Make this null target to left and right cameras. Now u have convergion point.

Here is the example. (http://rapidshare.com/files/112927022/Stereoscopic_Rigs.rar)
The first rig is the simple rig that I'm using everyday (I'm also in stereoscopic business), the second - same rig with sliders like u wanted (hope u know how to work with sliders in LW, personally i don't like using sliders at all, sometimes they make problems with network render ). U can animate the centrall camera (just us it for preview of your animations), or camera parent. Then u need to render from each camera (first left and then right one) to get stereo picture (obviously u need Anagliph Stereo : Compose filter enabled in image filters). U can move only Right camera only in X direction, the left one will move automatic.
Good luck.

[Varinder]

Hey Eugeny,

Thank you so much for these rigs, they were exactly what I was looking work. They work perfectly. Especially the convergence point NULL, this will tell me exactly how far a particular object is from the camera to determine the best interaxial distance. I mainly needed this feature because I needed to do a number of tests to understand stereoscopic cinematography better.

I’m also glad you noted that you have to ‘some times’ toe-in the cameras due to the left/right edge issues (although I’m not very familiar with this issue, maybe you can elaborate on that Eugeny). Because most or all of the time its best to have the cameras perfectly parallel with the convergence point at infinity. From what I’ve researched and have heard many experts in this field say is that toeing-in is a big no-no.

Reason for that is it creates vertical parallax known as keystoning. This would distort the image and make for an uncomfortable 3D effect. There should be zero vertical parallax. Eugeny, I’m guessing you know about all this already because it looks like you have a lot of experience with stereoscopic work but I thought I would re-explain certain things about stereoscopic cinematography for newcomers who are interested about it.

I thought I would mention an interesting fact about James Cameron’s 3D film Avatar and to further explain why ‘toeing-in’ the camera should be avoided. As some of you may already know, all of James Cameron’s 3D films were filmed with the ‘toe-in’ technique including Avatar. The filmmakers didn’t realize until later on that there was a huge amount of the keystone effect in their footage so they had to do a lot of extra work in post to fix the keystone problem. If it had originally been filmed with the cameras being parallel you wouldn’t have to fix any keystone issues in post, saving you time and money.

Another thing I would like to mention and I’m not sure if this is common practice in the stereoscopic business or not because I’m still new to it as well. I think most people already know that when you are shooting huge landscapes the interaxial distance has to be pretty far apart but when you are shooting normal scenes with people, the interaxial distance is much shorter. But, even when you are shooting a normal scene with people, you still have to continuously adjust the interaxial distance.

For example, when a character is at the end of the room and the character begins to walk towards the camera. The interaxial distance has to shorten as the character gets closer and closer to the camera. Same principal applies when an object is close to the camera and begins walking away, the interaxial distance has to go farther apart as the character travels further into the shot.

A lot of these methods I’m explaining are meant for real cameras that I’ve heard from experts in the field. I’m not sure if you can cheat a few things when it comes to CG shots when working with virtual cameras. Just to avoid future confusion, the 3D film that I will be working on soon is a live-action film, not CG. I’m just noting that because I’m still not sure if there’s any difference between live-action films shot with real cameras and CG films shot with virtual cameras in terms of stereoscopy.

I should also note that the 3D film I’m working on will be viewed with polarized glasses not anaglyph. But at home I have a 3D-ready Samsung DLP TV with a pair of i/o shutter glasses that lets me see awesome 3D images with no ghosting. And this is the way I’ll be viewing all of my content that I’ll be creating through lightwave.

Anyway, thank you again Eugeny for those rigs, I’ll be using that setup a lot from now on as I continue working on my own stereoscopic content.

By the way, it would be nice if everyone can continue to discuss stereoscopic cinematography in this thread to make it convenient and informative place to learn about stereoscopy.

Cheers,
Varinder

[Danimator]

Hello, let me see if I can shed some light on this.

3D in Lightwave and camera rigs can be fairly simple, you just have to follow some rules. Your best bet is to use the plug in above. Ive tested it and I'm pretty happy with it so far, mainly for its ability to render with Fprime, and rendering both views at once is a rendering time saver over rendering two separate images. Then in the comp you can split the render into two streams. If you want to use Lightwave's own stereo rendering, be aware that it will create 2 parallel cameras, or toe in cams if you target the camera to a null and limit its rotations to only Heading. If you rig a multi node camera, be sure to check its gimbal rotations. a stereo camera must always rotate and move as a unit. Some sloppy rigs Ive seen on the net will lock up, or rotate while maintaining the cameras up vector. Not good.

Now, I need to address the notion of animating the Interaxial camera separation.
THIS IS A BAD BAD BAD IDEA. As a viewer try to imagine how you would feel if the eyes on your head shifted in place from 3 inches apart to 10 inches apart. You would have a headache and feel dizzy. This is what animating this value within a shot is equivalent to and is almost never a good idea.

Toeing in your cameras only happens when your eyes converge on something that is close to your FOV breaking the screen plane. Anything that goes in -Z depth space behind the screen should never have key stoning or toe-in. In CG thats easy cause you can render the background with parallel cameras and the FG with toe in. In Live action toe in has to be fixed in post, or you can suffer from uncomfortable viewing. The foundry is working on a great plug in that helps to reduce these despairities.

I also use the Samsung DLP to view my 3D films. Its truly the best method out there for the money right now. I should also mention that I am creating a 3D stereoscopic training series that will expound upon these topics and more. Check out my website for more details

[Varinder]

Hey Daniel,

First off, really great post it was very informative, thanks for sharing.

About what you said about animating the interaxial distance. From what I’ve researched and have heard from a few experts in the field. They have explained that when a character is far off in the distance and begins to walk towards the camera, they would have to adjust the interaxial distance just like you would when you are adjusting the focus on the camera. I remember listening to a podcast from the upcoming 3D film Dark Country and the filmmakers also noted that they have done this on set while filming. I haven’t personally tested this out yet and seen the effects myself, but at the moment it makes sense and it would make for a more comfortable 3D experience.

For example, say if we had a character 30’ away from the camera and the interaxial distance is 1’ away (using the 1/30th rule as a general rule of thumb) and the character walks right up to the camera now only 5’ away. Of course this would have to be a fairly long shot for the character to walk that far. But when the character approaches the camera with the interaxial distance 1’ away, it would be fairly hard to fuse the 2 images together with our eyes since they would be so far apart. So instead when the character is walking towards the camera up to 5’ away, you would slide the camera’s closer together to about 2” a part (again, using the 1/30th rule as a general guideline) for a more comfortable viewing.

Again, this is something I’m only guessing would happen since I haven’t done any of my own tests yet. Perhaps you can simply animate the Zero Parallax Distance/Plane (ZPD) in the shot so the audience has a better chance of fusing the 2 images together as the character comes closer to the camera. But that might lead to too much positive parallax.

Maybe someone with more 3D experience than I do can elaborate more on this situation.

By the way Daniel, I’m really looking forward to your 3D training series, do you have an idea when you might release it?

Cheers,

Varinder

[Danimator]

I strongly disagree with any "expert" who thinks moving the interaxial distance in that matter will provide a better experience. This topic is debated but there are some fundamental truths you need to understand. 3D stereo is not a gimmick or device. Its not rack focus, jump cuts, or some cheap trick.

3D is the way humans see the world. Binocular vision, and the way our brain interpolates the information on a human level is the most important consideration when presenting good 3D that will not cause eye strain or discomfort for the viewer.

To this end, dynamic adjustments to the interaxial is making your brain feel like your head is expanding and contacting in scale when relating to the picture plane. This is not natural and inherently wrong. While this can be used for certain effects, like I did for an amusement park ride where the viewer was being magically shrunk in size. Its never a good idea for 99% of 3D imagery. It also will cause the scale of the projected image to change radically within a shot. If a person is walking towards camera and your interaxial changes from 6 cm to 2 cm the 3d image will appear to distort and the scale will appear to grow or shrink.

Try thinking of it this way... If you have the pleasure of being on a film stage with a miniature photography unit, why do miniatures look so small? When you photgraph them in 2D and project them, they look so real and they appear the proper scale. This is because your eye separation of 65mm tells you that your looking at somthing small. If they filmed that same miniature for a 3D film, then they would have to use a beamsplitter and get a interaxial separation of not 65mm (cause that would give you the same scale as real life) but that of a human scale for the miniature. So if the model is 1/8 scale then your interaxial will need to be 1/8 of 65mm so your seeing through the eyes of a scalled down person seeing at that level.

I hope this clears it up.

My Training series will start rolling out in July. I'm going to be covering allot of material and techniques for almost all software, from Maya to Lightwave, Fusion to Nuke.

[seriphu]

There has been talk at work about doing test implementation of stereoscopy on some projects. I understand anaglyph fairly well, however polarized stereo is where I start to lose focus. What is the difference in camera shooting other than the polarized lens/filter? After the film is shot in either 3D software or on set is there a specific projector or monitor that must be used to show the polarized movie?

Thanks guys

[Danimator]

Well the thing your missing is there is no difference for acquisition of media for the method of Stereopsis. In other words, you do not do anything different.

All 3D imagery starts with a Left eye render or camera and a Right eye. Once you have that, the method you use to display the 3D to the viewer is what is different. Polarized, Interlaced, CheckerBoard, Anaglyph, ColorCode, RealD, IMAX 3D, or Dolby 3D, not any of them matter in how you acquire your footage. Now screen size, where you sit in the theater, and the amount of camera separation (Interaxial) do matter quite a bit.

The second part of your question is how do you see a polarized movie. Well you will need to setup two synchronized projectors or one RealD z-screen. Then you add polarizing filters to the projection lens, use a non polarizing silver screen, and use the matching polarizing glasses to decode the image for each eye.

[seriphu]

Thanks so much Danimator, that clears up a lot. Is it possible to view polarized stereo on a television? I was told that it may be possible with a lenticular screen, however I have not heard of "lenticular" before.

Do you happen to know of any reading material that would be a good place to start for the basics of stereo production and beyond?

Thanks again mate

[Cheesestraws]

This whitepaper written by Autodesk is probably one of the best resources for information on stereoscopy in production.

http://area.autodesk.com/images/tips/the%20area/080911_2145/stereoscopy_whitepaper.pdf

[icester]

This whitepaper written by Autodesk is probably one of the best resources for information on stereoscopy in production.

http://area.autodesk.com/images/tips/the%20area/080911_2145/stereoscopy_whitepaper.pdf

Missing link!

All those pseudo-stereo theories...

CG Camera geometry must match the end user/viewer stereoscopic view geometry and not the other way around.
If there is a mismatch than the objects of a scene are squeezed or stretched in depth and vergens is only possible on the partial 3d volume.
That is all to it.

In practice, based on real data form Cinema theatres and HOME setups one can design and simulate use/viewer stereoscopic geometry .
Ones determined such geometry reference can be used as a template for creating stereoscopic content with matched stereo camera geometry.

The preferred geometry for cinema and HOME theatres is parallel.
This allows viewer to focus and converge on scene objects both in center any away.

The converged stereo camera geometry can only be used with HMD's which have build in eye tracker for dynamic vergens point matching.
This type of geometry give most natural stereoscopic experience but it is very expensive and limited to singe viewer per system.

Thank you for posting the links to those stereo camera plug-ins.

Mathew Orman

[http://www.tyrell-innovations-usa.com/] (http://www.tyrell-innovations-usa.com/%5D)