Camera Converged or Parallel?

[Rich Schaefer]

I have been a believer in shooting in camera converged and I have been doing so, but I think I am becoming a believer of shooting parallel. I wanted to see what the other shooters of stereo were thinking these days?

I should also add that John Schwartzman shot parallel on the current Spiderman and I thought the stereo was very well done and most of all very easy 3d to watch.

Below are the pros and cons of each in my mind. I see there are two reasons to shoot converged and 3 good reasons to shoot parallel

Pro Converged arguments:
1. There is a little, perhaps 5%, more perceived roundness on objects but you really-really have to be looking for it to notice it.
2. Its easier to shoot converged because it displays 3d well on the onset monitors unless you use a convergence box like a Davio.

Pro Parallel arguments:
1. I think it is an easier to view and resolve the 3d, because the sides work/fuse better because the cameras are pointed in the same general direction.
2. In post there is no Key-stoning to correct like there is in camera converged meaning potentially easier/faster post production.
3. Due to the nature of optics being very good but not perfectly square, keeping the anomalies of bend and barrel distortion of the two images pointed on the same points will make for a better match once blended in to one stereo image.

Here is a test I did, one is camera converged and one is parallel and converged in post . They are maybe 15 seconds apart with the same IO. The red letter on the left "C"=Converged and "P"=Parallel. Check them out here on youtube 3D:
http://www.youtube.com/watch?v=8kAFnfOE0_g

What do you guys think these days?

Thanks much!
Rich

 

[Dave Blackham]

Pro Converged arguments:
1. There is a little, perhaps 5%, more perceived roundness on objects but you really-really have to be looking for it to notice it.
2. Its easier to shoot converged because it displays 3d well on the onset monitors unless you use a convergence box like a Davio.
h

There is potentially more resolution with converged as there less crop applied from the HIT transform with parallel filming.

Pro Parallel arguments:
1. I think it is an easier to view and resolve the 3d, because the sides work/fuse better because the cameras are pointed in the same general direction.
2. In post there is no Key-stoning to correct like there is in camera converged meaning potentially easier/faster post production.
3. Due to the nature of optics being very good but not perfectly square, keeping the anomalies of bend and barrel distortion of the two images pointed on the same points will make for a better match once blended in to one stereo image.

Im not sure I agree with 2. As you mostly not converging that much in camera that keystoning is an issue in post.

With both techniques/tools exist in post for good image alignment for Converged or Parallel filming. Both techniques are equally valid depending on the sequence in question. The most important thing is matched lenses and rig alignment.

 

[Rich Schaefer]

There is potentially more resolution with converged as there less crop applied from the HIT transform with parallel filming..

True, I did have to blow up the in camera converged so they would appear the same size. It was by 2% that I zoomed in. I would shoot at least 4k for stereo to have extra pixels for post fixes.

I'm not sure I agree with 2. As you mostly not converging that much in camera that keystoning is an issue in post.

I recently had a great series of steadicam shots we did that were shot side by side rig, camera converged, shot with 11mm lenses into a stadium with lots of detail. I spent some time with those shots in post and the key-stoning was severe. It was those shots were it started to occurred to me that I would not have this problem if the cameras were parallel.

 

[Pawel Achtel]

Rich, you are absolutely right.

It is maths, or specifically geometry and common sense that the entertainment industry has been historically slow to adopt.

It can be argued that the resolution loss resulting from correcting the key-stoning is larger than that from HIT. But it is difficult to correct the asymetric geometric distortions in left and right eye as the lenses rarely render the field flat and undistorted. And, more importantly, when shooting converged, the image planes are no longer parallel, so that the focus on each side of the screen falls at a different distance on left and right eye. This is not assisting the brain to fuse such images together and can not be "corrected" or compensated for in post.

Convergence (toe-in) gained popularity in live 3D broadcast as a "cheap", or rather gimmicky, way of adjusting the depth at which the image appears to the viewer. Complex systems, like CPG Fusion 3D, have been developed and marketed convergence linked to focus as a desireable feature, for example. Hype over science. It happens all the time.

On a side but related note, I know of some clowns advocating convergence when filming underwater through a flat port without understanding that it works like a variable chromatic aberration and asymetric distortion control allowing the user to change them from "severe" to "ridiculous" and anything in between. Welcome to the entertainment industry! :banghead:

 

[Dave Blackham]

Rich, you are absolutely right.

It is maths, or specifically geometry and common sense that the entertainment industry has been historically slow to adopt.

It can be argued that the resolution loss resulting from correcting the key-stoning is larger than that from HIT. But it is difficult to correct the asymetric geometric distortions in left and right eye as the lenses rarely render the field flat and undistorted. And, more importantly, when shooting converged, the image planes are no longer parallel, so that the focus on each side of the screen falls at a different distance on left and right eye. This is not assisting the brain to fuse such images together and can not be "corrected" or compensated for in post.

Convergence (toe-in) gained popularity in live 3D broadcast as a "cheap", or rather gimmicky, way of adjusting the depth at which the image appears to the viewer. Complex systems, like CPG Fusion 3D, have been developed and marketed convergence linked to focus as a desireable feature, for example. Hype over science. It happens all the time.

On a side but related note, I know of some clowns advocating convergence when filming underwater through a flat port without understanding that it works like a variable chromatic aberration and asymetric distortion control allowing the user to change them from "severe" to "ridiculous" and anything in between. Welcome to the entertainment industry! :banghead:

It rather depends on the sequence of course as to which technique you wish to use. Both are used to good effect when done properly and used by professionals all the time. Theres much more to 3D than sharpness of image, Variable IA and convergence including the ability to work to small IA are essential to the normal in air environment as well as Underwater of course.

 

[Pawel Achtel]

David, this thread is not about IA. We are discussing convergence (toe-in).

Can you explain how convergerce produces good effects "when done properly" or "used by professionals"?

I don't think geometry or physics change depending on who is doing it. :hand: You either converge or shoot parallel and do HIT.

What do you mean by "done properly"? Rich and I believe that we should always shoot parallel and both Rich and I put strong arguments for it.

You are also incorrect about sharpness disparity between left and right eye, which is what you will always get when converging. It is bad. If you think the same focus plane for right and left eye are not important, please explain why because I'm not getting your point of view.

While you can get away with toe-in on long lenses, I don't understand why convergence is desirable or "essential", as you write.

I believe that convergence is a form of tax on cinematographers that do not understand geometry.

 

[Peter Lyons Collister ASC]

On Spiderman John decided to go parallel on good advice from Rob Engel. We did allow for crop but we were Epic so that was easy....Viewing video assist in Parallel was no problem and convergence is always manipulated in post no matter what you do in production...

In my mind it is a win-win situation to shoot parallel IMHO

 

[Pawel Achtel]

Thanks. Totally agree with you, Peter.

 

[Rich Schaefer]

Thanks for the replies everyone! It sounds like I am not alone on my thoughts about shooting parallel.

Pawel,
I totally agree, all of your arguments make sense. I don't know how anyone can think shooting toe-in under water would be better. I would sure think keeping the light as parallel through the port would lend to less light bending, fringing and aberrations...

Peter, thanks for the confirmation, and yeah, I agree, win-win!

 

[Rich Schaefer]

Whats everyone else doing, camera converged or parallel?

 

[Jason Goodman]

When I first started shooting 3D in 1991, I was adamant about crossed lens axes (converged cameras) because that's what I learned from Foundations of Stereoscopic Cinema. Over time I've changed my opinion and now prefer parallel. BUT WHEN SHOOTING PARALLEL YOU MUST HAVE HIT ON SET IF YOU WANT TO DO IT RIGHT. Many people forget this step, but that's like pulling focus with your eyes closed.

When we created the 3DVX camera system, we developed optical horizontal image translation (OHIT) which offers the best of both worlds. Real time in camera zero parallax setting without resolution loss, keystoning or physically moving the cameras. OHIT has been incorporated into the Panasonic 3DA1, 3DP1, Canon XF305, XF105 and Sony TD300.

There are still times I prefer converged however. Parallel is great for RED because you've got so much resolution, using anywhere from 2% - 10% or more for stereo rectification does not have a significant impact on final image quality, particularly if you're delivering HD or 2K. When shooting live, or shooting 1920x1080 to deliver 1920x1080 converging in camera is best. I also do it when I know I am handing off footage to a post house that is not experienced in 3D. There have been times I've shot parallel and not been in on the 3D DI and come to see my work completely not rectified, and that is awful.

 

[Maciej Orman]

I suggest that the best way is to use neither.
Instead use off axis geometry by implementing Canon's shift lenses.
In such configuration you simply set the stereo window to the target screen size and you get perfect, realistic stereoscopic content which requires no post editing.

 

[Pawel Achtel]

Is it equivalent to shifting the sensor horizontally using regular lens?

 

[Jason Goodman]

it's an interesting question, but no they are not equivalent. Shifting the entrance pupil would actually change your interaxial. Using the OIS mechanism (OHIT) or moving the actual sensor would not and are therefore the preferred methods.

 

[Maciej Orman]

Is it equivalent to shifting the sensor horizontally using regular lens?

Yes,
except when lenses are shifted it mimics natural translation of eye's entry pupil thus when setting stereo window the geometry is matched with human vision system.

 

[Rich Schaefer]

Jason, you are right on with your thoughts on 1080 cameras, live TV and non-3D savy post houses! Great points to consider.

Meciej, the idea of using the canon tilt-shift lenses is very interesting! Someone should test that, but the alignment, could be a real bear

 

[Dave Blackham]

David, this thread is not about IA. We are discussing convergence (toe-in).

Can you explain how convergerce produces good effects "when done properly" or "used by professionals"?

I don't think geometry or physics change depending on who is doing it. :hand: You either converge or shoot parallel and do HIT.

What do you mean by "done properly"? Rich and I believe that we should always shoot parallel and both Rich and I put strong arguments for it.

You are also incorrect about sharpness disparity between left and right eye, which is what you will always get when converging. It is bad. If you think the same focus plane for right and left eye are not important, please explain why because I'm not getting your point of view.

While you can get away with toe-in on long lenses, I don't understand why convergence is desirable or "essential", as you write.

I believe that convergence is a form of tax on cinematographers that do not understand geometry.

Pawel, Given the option most of the time we would prefer parallel but there are times when we would need or want to shoot converged. Long lens work being one at times, though for obvious reason we steer clear of long lenses for 3D, macro being another again at times. Both techniques are valid and are used, it doesn't mean one is right or wrong it depends on what is needed to be achieved. Respectfully, I just disagree with the way you phrased you post.

 

[Pawel Achtel]

Both techniques are valid and are used, it doesn't mean one is right or wrong it depends on what is needed to be achieved..

Dave, I made an assumption that what needed to be achieved was "good 3D" and ignored other considerations. Convergence is bad because it creates image focus plane disparity between left and right eye and because it introduces key-stoning which, as Rich has rightly reported, are very difficult to get rid of when taking into account other distortions of the lenses.

I also agree with Jason that HIT should be done on set to find the best IA for each shot.

Jason, interesting idea with OHIT (optical horizontal image translation). I was wondering about that and even started designing a modified lens mount for Epic to do so, ie, shifting the lens relative to the sensor. It would only require tiny amount of travel (sub 1mm) to cover reasonable range of zero parallax adjustment.

BTW, with 5k or 6k resolution to spare, HIT may be practically implemented incamera firmware by windowing the sensor. How about this as a feature request to RED?

 

[Rich Schaefer]

BTW, with 5k or 6k resolution to spare, HIT may be practically implemented incamera firmware by windowing the sensor. How about this as a feature request to RED?

I like this idea, I vote for this option.

What are you guys using for on-set HIT? Davio? Do Re Me? Transvideo or Panasonic 910 monitors?

 

[Pawel Achtel]

I use Marshall OR-70-3D due to the size limitations, but it does the job.