POLL: Should stereo cameras point inwards, or be parallel?

[phil]

[cybereality]

I'd say in general to go with parallel. If you were say filming a big-budget movie, then maybe you'd have the luxury of adjusting the convergence for each shot. But unless you are doing this every time then the convergence is mostly going to be wrong and you are better off with parallel.

[Okta]

So im the only one who's right so far? Toed in is the correct way.

[WheatstoneHolmes]

IMO: go with whatever looks good to you.

I used toed-in on my "S-3D Shenanigans #1" http://www.mtbs3d.com/phpBB/viewtopic.php?f=98&t=3713 for some scenes and parallel
for others.

The bubble blowing scenes are toed-in.

I did not vote because I don't think that one can absolutely use one over another in every situation. *steps off soapbox*

[Freke1]

I read some of Andrew Wood's papers and toed-in creates some distortions (misalignments in heights) that parallel cameras doesn't. However I saw a picture of a big Hollywood 3D camera (Cameron's?) that used/could use toed-in.

[Neil]

Toed in is definitely the wrong way to go. It creates distortions.

Better to shoot with a little extra field of view, and adjust convergence afterward.

Regards,
Neil

[Okta]

Toed in is definitely the wrong way to go. It creates distortions.

Better to shoot with a little extra field of view, and adjust convergence afterward.

Regards,
Neil

The brain corrects the distortions. Youe eyes naturally toe in on whatever you look at so there is no way around it. Its normal.

[cybereality]

Your eyes naturally toe in on whatever you look at so there is no way around it.

That is exactly the problem. If you film like this then you are telling the viewer: "you must focus on this one specific point on the screen and nothing else". If the viewer were to look anywhere else in the image it would be weird and not correct. This may be acceptable for a very simple scene, like a product shot on a white background. But if it were a complex scene it may not be possible to know exactly where the viewer is going to be focused on at any one moment. In that case parallel would be the superior option.

[Okta]

Good point, you got me there

[AWoods]

This is an often misunderstood topic so it is useful that it has come up again.

Firstly, there are actually Three basic stereo camera setups:
1. converged via inward rotation (toed-in)
2. converged via image shift (sometimes called 'parallel' or more correctly 'parallel with image shift')
3. parallel (without any convergence or image shift)

Toed-in cameras produce keystone distortion and depth-plane curvature.
..(explained here: http://3d.curtin.edu.au/spie93pa.html" onclick="window.open(this.href);return false; )
..Keystone distortion can produce eye strain - particularly in the corners of the image.
..Depth-plane curvature produces a warping of 3D space which may not be perceptible.
..Despite what Okta said that "the brain corrects these distortions" - this is not correct.

Option 2 (converged via image shift / parallel with image shift) avoids these distortions
..However it can be hard to implement.
..It either requires physical modification of the cameras (shift the CCD behind the lens) or
..shifting of the images in software (but with the disadvantage that some of the sides of the image must be cropped).

Option 3 should be avoided since objects at infinity will be placed a screen depth, and everything else will come out of the screen. This could be fixed by shifting the images, so that the 3D image is shifted back in depth, which gives us Option 2.

In the ideal world Option 2 would always be used - but Option 1 is often used because it is easy and expedient.
If the video shot with a pair of toed-in cameras is only going to be shown on a small screen, the distortion and resulting eye-strain may not be noticeable.
It is theoretically possible to remove keystone distortion by further image processing, but this function isn't available widely. ("StereoPhoto Maker" is one piece of software that can do this).

So, in conclusion, toe-ing in a pair of web-cams is probably going to be OK if it is only shown small screen, but if you're aiming for high-quality, Option 2 (converged via image shift / parallel with image shift) will be better.

How does this explain why Pace use toed-in cameras?
Just watch Hannah Montana 3D or Jonas Brothers 3D and you'll see the distortions I'm talking about.

[Okta]

My right eye was damaged long ago with retinal tares and i ended up with distortion in my right eye vision that curves or bulges straight lines. My left eye is still good. When i look at anything with both eyes all lines appear straight. This is teh magic that the brain does for our vision. I also tried a side by side mirrord s3d setup using 1 16.9 monitor and 1 4.3. It worked because our brains and eyes make up the differences.

[Likay]

Toe in is NOT the correct way to record stereo-3d. However: When pictures from a stereocamera is viewed "as is" in a stereoviewer, everything appears in front of the screen. This is best adjusted using image-shift. That way you can also use the convenient batchconvert function in s-photomaker (when setuped right all the pictures will have a good aligning without any more job). Using toe-in (especially if not constant) you need to align all pictures one by one. There is stereo in a a toe-in video but it's very easy to get eyestrain because of it, and even more hurtful viewed on a bigger screen.

[BlackShark]

Neither of the two proposed ways.

Pure parallel is wrong because everything pops-out, guaranteeing headaches.
Pure toe-in is also wrong because : 1- it's very easy to screw the shot with too much toe-in, 2- introduces vertical misalignment towards the edge of the images (depends on the amount of toe-in)

Asymmetric frustum (i think that's the name) is the only correct way to do it but it requires shifting the lenses which is absolutely impossible to achieve on a standard consumer camera... and even all pro cameras i think. You'd need a real custom rig to do it properly.

Parallel is easy to correct on post-process, but not everyone will accept to loose image definition, and/or do not want to post-process at all.
Toe-in is harder to correct on post-process, but if you don't film in close-up you'll never see the deformations. So it's much more suitable for point-and-shoot full automatic amateur filming. So I'd recommend toeing in at a fixed calibrated angle.

[DavidGhast]

Pure parallel is wrong because everything pops-out, guaranteeing headaches.

What makes you think parallel doesnt produce depth into the screen? Are you saying the engineers over at red cinema, as well as everyone else whose ever made a 3d rig since the very beginning are wrong?

[BlackShark]

Pure parallel is wrong because everything pops-out, guaranteeing headaches.

What makes you think parallel doesnt produce depth into the screen? Are you saying the engineers over at red cinema, as well as everyone else whose ever made a 3d rig since the very beginning are wrong?

Mathematics... and about every single web publication about "how to shoot stereo".

Pure parallel (no sliding in post) does not produce depth, but it is easy to correct in postproduction and much easier to build the camera and to shoot. That is only if you intend to edit the footage and accept to loose image definition.

[yuriythebest]

Pure parallel is wrong because everything pops-out, guaranteeing headaches.

What makes you think parallel doesnt produce depth into the screen? Are you saying the engineers over at red cinema, as well as everyone else whose ever made a 3d rig since the very beginning are wrong?

Mathematics... and about every single web publication about "how to shoot stereo".

Pure parallel (no sliding in post) does not produce depth, but it is easy to correct in postproduction and much easier to build the camera and to shoot. That is only if you intend to edit the footage and accept to loose image definition.

I think the sliding in post"+ parallel is he best and most reliable way to get good 3d- almost everyone uses it and it seems to yield the best and most consistent results, all be it with perhaps some loss in resolution due to the areas that were "slided away" The red cameras, the minoru3d, the upcoming fujifilm 3d camera and all/most of the ancient film based 3d cameras use parallel

[phil]

Sorry, I've brought us into another terminological tangle (like the one we had with the term "convergence", which I learned can mean about fifteen different things).

I didn't realise that the term "parallel" could sometimes imply a lack of horizontal image translation.
So the poll was purely about the physical angling of the cameras - I was treating it as understood that with parallel cameras, the images would need to be separated one way or another (by offset camera sensors / post adjustment / asymmetric frustum).

cheers 8)

[Okta]

Perhaps it all comes down to the display method. If you have a life size dislplay method where the centre of screen is taken up with the subject matter toe in will be best. You really should not be looking around anyway. This will give the most natural result and feel like proper 3d instead of a weird effect.

[DavidGhast]

What is the difference between "sliding" during production (i assume by this you mean adjusting the interocular distance) and sliding in post, aside from the loss of quality incurred by post processing? Interocular is interocular.

[yuriythebest]

What is the difference between "sliding" during production (i assume by this you mean adjusting the interocular distance) and sliding in post, aside from the loss of quality incurred by post processing? Interocular is interocular.

the image will look a tad different when you actually slide the cameras or adjust their convergance in real life versus sliding n video post producion- you might not notice the difference but the part of your brain that has to fuse em into the 3d image sure does

[phil]

Perhaps it all comes down to the display method. If you have a life size dislplay method where the centre of screen is taken up with the subject matter toe in will be best. You really should not be looking around anyway. This will give the most natural result and feel like proper 3d instead of a weird effect.

Hi Okta,
can you explain why toe-in is the better method in that example?

Why would parallel cameras + image shift not be suitable for that case?

In what cases do you think parallel cameras + image shift would give the best results (and why)?

[cybereality]

Ok, I am a little confused here. Last time I checked, my actual eyes are toed-in just a little bit when I focus on something. Does that not produce distortions in my brain?

[Tril]

I think that the thing is that you toe-in your eyes when looking at your 3D monitor. The amount of toe-in of your eyes changes as you look at different depths in the picture. If you toe-in the camera, your eyes are still free to look anywhere on the screen and change how much they toe-in. Maybe that causes visible distortion. That's my guess on the topic but I don't have hard evidence to prove my claims.

[BlackShark]

Toe-in distortions are not related to the way eyes work. It's related to how the screen displays the captured images. The distortions can be corrected in postproduction but the actual software for that is more difficult to find (i guess it's a specialized and very expansive software).

[yuriythebest]

yeah- I think you'll get "double toe in"- since your eyes are already toed in, plus the toe in of the camera- which would look weird - or something like that- could be wrong

[Okta]

Toe-in distortions are not related to the way eyes work. It's related to how the screen displays the captured images. The distortions can be corrected in postproduction but the actual software for that is more difficult to find (i guess it's a specialized and very expansive software).

Can you go into further details on that? I am not understanding the difference between toe'd in eyes and toe'd in camera's.

[cirk2]

Ther is no difference between toed-in Cammeras and Toed-in Eyes... If you project the image of the cammeras direct into your brain...

If you look at the Toed-in Picture with your toed-in eyes your eyes have to be more seperated to get the same effect.


Just thing about triangulation, your eyes/ the cams are two points the object is the third...

[gzhang207]

Parallel configuration gives the simplest math form in getting 3D from disparity, but it suffers from the problem of small overlapped areas on camera between left and right. Pixels that can't be fused into 3D can confuse the eyes. Also the use of wide angle lens to compensate actually makes the unmatched optical distortion worse.

The toe-in configuration via convergence is often preferred for maximized overlapping between the left and right eyes. However, the angle depends on the sweet spot of 3D scene. One commercial stereoscopic imaging device has 3 deg half convergence angle with an optimum view depth of 20cm.

The human binocular vision works well within an arm's length and beyond that it is based on knowledge via reference. The same applies similarly to S3D.