MTBS Stereoscopic 3D Settings Guide

[Neil]

Hello everyone!

I took a stab at a stereoscopic 3D concepts guide, but I haven't made it an official addition to the site yet because I think the diagrams can still be better, and I'd like to get a sense of what the members think of it. I'm happy to revise and adjust accordingly.

You can read the RC1 version here:

http://mtbs3d.com/cgi-bin/rss_blog.cgi?news_id=44" onclick="window.open(this.href);return false;

Regards,
Neil

[LukePC1]

Well done. With nice pictures and examples.
Unfortunatly the movies - they are movies, right? - don't work for me with FF. I haven't tried IE.

They are 'just' a picture after a long loading time. I hope you can correct that somehow

[Neil]

Do you see a play button? You need to click the picture to get it to run.

Regards,
Neil

[pixel67]

I am using firefox without a problem.

Great job on the content, Neil.

[Neil]

There might be a media player update plug-in for Firefox. Sharky knew about it.

Regards,
Neil

[stee1hed]

Excellent job Neil, I learned something!

I used Firefox and it worked fine. I right clicked on the movies and selected play.

The examples were right on and what pushes this beyond just an average tutorial. I look forward to others.

Suggestions... it's late and I can't think of any at this point.

[stee1hed]

I just thought of something! It might be cool to link to that one site that shows people how to test their 3d vision with their fingers or to create your own test for readers to try in the tutorial that reinforces your points.

[Neil]

Give me some links, and I can try. Do you think the images are public domain?

Regards,
Neil

[Tril]

I get the problem with the videos too on Firefox. It shows a black square in which it's written "(no video)" (without the quotes) and there's no play controls below the black square.
The videos work fine on Internet Explorer 7.

The content is good. I think it's well explained.

EDIT : I got the videos working.
I found this : http://kb.mozillazine.org/Windows_Media_Player
This lead me to : http://port25.technet.com/archive/2007/ ... refox.aspx
Which gives a plugin for videos on this page : http://port25.technet.com/pages/windows ... nload.aspx
I installed this plugin and now the videos work.

[LukePC1]

Ok after reading here again, I used rightclick and play/stop (or something similar) and now it works. Thank's for the hint

[InCytE]

Freak'in excellent Neil. Hardware manufactures should be including this with all their S3D solutions.

[Neil]

I still want to do some revisions here and there. I got some good feedback that the introduction can still illustrate some common problems that cause bad first impressions, or something that better illustrates why good settings are important.

I could use some suggestions on a follow-up guide. I figured I'd write usage guides for the iZ3D and NVIDIA driver solutions. Maybe even an anomaly diagnosis system.

Can you share your first experiences with your first S-3D solution, and I can work on the guide to acknowledge them (e.g. I wish someone told me about...).

Regards,
Neil

[stee1hed]

Here's the site where you can test if you have 3d vision

http://www.vision3d.com/frame.html

[Sabre2552]

Maybe you could describe how to get the most realistic S3D experience?

[Mechangel]

I am using Firefox, videos loaded immediately and had no trouble playing them.

I am a 3D noob and that page really helped me understand how the different types of 3D work.

I am confused which type of glasses one would use to view them though? I haven't bought any glasses yet.

[LukePC1]

I think you can't actually watch the videos in 3D. Their purpose is more to show, how it should look WITHOUT the glasses. If you put on the glasses after you have configured it like the examples it should have the described effect.

The advantage of the videos is, that everyone can watch them and get an Idea, of how it should look like without the glasses.
Unfortunatly a lot of reviewers have different pictures up, so you don't realy know, where the 3D is coming from - of if the reviewers had the right settings.

I think there were reviews out where testers didn't experience any 3D, because of wrong settings. The result is then: don't get it, 3D is not that good after all...

As soon as you got your glasses you should understand better. Or maybe try to make something similar with anaglyph (Red/cyan) glasses...

[Mechangel]

That makes sense! I ordered the transmitter and active glasses last night so I'm excited to try it.

[digitalmellon]

Hey Neil, i don't get it.
This "Depth Only Settings" theory especially.
Before i bought glasses, i thought it is only one theory. Because we are build one way.
2 eyes, perspective. Now i see some crazy theories like "Depth 3D" and "Pop-out 3D". What the hell?

Whole respect man (all of you guys),..
"you can see into the screen like a car windshield", "seem to come out of the monitor"?

"Pop-out effect" - i thought it's normal life perspective. Nothing "comes out" My Monitor.
It just like normal binocular vision - strait lines into vanishing point.

Feel like I'm lost.

P.S. I think u have took whole "Depth 3D" theory for granted. I'm pretty sure what i'm saying, but don't have means to proof that yet.

[Likay]

Not sure about what depth3D means but i think it's more or less 3d-vision over all. Pop out is simply objects in front of the actual screen. For example if something comes flying towards you, continues through the screen until you more or less feel it passing through your head! Usually when playing games it looks best playing with little pop out (depending of what kind of games of course) but fps's and adventure games works ok with this basis settings. Even so, some things seems to appear very close to your head and even feels like passing through it. And my gaming screen is 4 meters away!

You have 2 basic settings for 3d in both the nvidia and the iz3d driver: Convergence and separation. Increasing separation increases the 3d-effect. Increasing convergence makes the whole gaming scene moves towards you. You also sometimes have another setting with which you can alter the screen size. This is useful for far away objects to appear... far away. Hope this might be helpful.

cheers

[Neil]

Hi Digitalmellon,

Thanks for reading the guide and sharing your feedback.

You are correct about your personal vision experience, having two eyes and seeing things from a fixed point with little to no differentiation between "depth 3D" and "pop-out 3D". This is correct because everything is done for you naturally.

In the 3D video game world, it's a bit different. Think of your 3D solution as a 3D sandbox where you can adjust the experience to your liking. Do you want to see deep inside the screen, do you want things to protrude in front of the screen glass, or do you want both?

The car windshield was an analogy of this experience. Did you want the bullet to hit the car windshield and stop there? Or did you want the bullet to go through the car windshield and seem to almost reach out and hit you? In the real world, you don't have these choices. In the 3D video game world, you can control this.

Depth 3D refers to the zone inside your monitor glass - the depth. Pop-out 3D refers to the zone in front of your monitor's glass - the pop-out experience.

Regards,
Neil

[crim3]

Keep in mind that this is all about a fake 3D. That's why there are several parameters that can be changed to match you sterevision in real life (in the firsts place, each of us have a different inter-pupil distance) or just your liking for each game, which is also an interesting feature.

One thing that real life doesn't have is a display that becomes a distance reference of the objects you see. That's why we talk about depth when stereo effect brings the illusion that the objects are further than the display surface, or pop-out when they are closer.

[digitalmellon]

There's only one reality i live in. Sorry.

The "depth 3d" settings you talked about is artificial, and brain can't take it easily.
In fact, having distant objects more separated than close one seen to brain like enlarged, because they should be closer considering perspective.
And that's all your sensation. That's why most people don't understand that - the brain reject it.

I have steped into topic (don't remember where) where some guy compared or tried to replace stereo by Fresnel lens in front of monitor. It was just the same to him.

If you want stereo image in real world, you have to use 2 parallel lenses contraption to take a shot. Parallel. And separated by a specific amount of space. Not necessarily "inter-pupil distance". If you go upper you get hiper-stereo (feel like a giant :] ). Bellow is use for macro. There are formulas for that.

The same should be available in games. Just one parameter - separation - and parallel cameras - no convergance.
Convergance maybe good for stills, but walking with zero paralax set in front of you is like walking with your finger in front of your face - and you can't look farther.


By the way. Neil we should revision your sketches 'cause they don't reproduce what exactly is taking place - take into account cameras/viewports into the sketches.



I'm all into popularizing stereo and taking it seriously as You.

Cheers

P.S. My english should be taken as is. ;>

[crim3]

In fake 3D convergence is necessary for geometrical reasons, though it's possible to implement zero-convergence stereovision if it's done natively in the graphics engine.

This is from the z800 SDK documentation.

Zero-Convergence Stereovision
Why do we need Zero-Convergence Stereovision?
There is nothing wrong with Stereovision as explained in the previous pages, also known as “Toe-in” Stereovision. It is very easy to understand and almost trivial to implement, has absolutely no impact on the game or application rendering performance and it produces very good and convincing stereo pairs: It is perfect in many situations.

However, end-users are more and more using their favorite games for hours and hours of fun. In the long run, after a number of hours of continuously playing games in stereo 3D, “Toe-in” Stereovision might cause some eye stress because of convergence.

The next picture shows the “Toe-In” fields of view from the top. In black, the non-stereo field of view. In red and green, the adjusted fields of view for the left and right eyes. It is obvious that both rays issued from each “eye” are converging to the same “look at” point. This is “convergence”. Of course, the effect is immensely exaggerated in the figure below for clarity.


Furthermore, what the users actually see in the game or the application, namely the projection planes, differ. They differ slightly when the separation parameter is small but they differ significantly enough to sometimes cause minor visual artifacts when separation is by a large amount.

The solution is “Zero-Convergence” Stereovision where the left and right view fustra become asymmetric to better match what is actually happening in the real world: our two eyes look ahead in the same direction, as shown in the figure below:

This solves two problems: There is no convergence at all any longer and both left and right eye projections yield the same projection plane. As a by-product, each eye now has its own “Look At” Point.

However, nothing comes for free, and we now have to implement projections with asymmetric view fustra. Asymmetric, because the field of view (also known as the aperture) is asymmetric: the aperture angle on the left of the right eye ray is larger than the aperture angle on the right of the right eye ray. The situation is inverted for the left eye ray.

Luckily for developers, both OpenGL and DirectX come “equipped” with the ad’hoc tool to let us easily create asymmetric projection matrices:

OpenGL features
glFustrum

and DirectX Graphics proposes
D3DXMatrixPerspectiveOffCenterLH.

Implementing Zero-Convergence stereovision is now a just bit more involving than the simpler “Toe-in” method but these API’s help a lot as we will see below in the programming samples.

I have to go right now, but I'll edit the post to add the pictures that come with the text. Anyway, the z800 SDK is available to download at www.3dvisor.com. The sections explaining the stereoscopic effect are worth the read.[/img]

[digitalmellon]

I just always thought that the eyes should converge on the plane, not cameras.

Hmm, i must think it up.
Right now i'm looking for some other sources too.

Thanks for the answer crim3.

I'll be back. ;]

[crim3]

Yes, the eyes converge for near objects and look straight, or almost, for far objects, but the objects must be at the right spot for this to be possible. I have to think about it throughly, too. I found out the need for convergence when I modified a direct3D sample to add stereo. With separation there was stereo effect but when I walk away from the objects they didn't look as being far because the left and right images didn't separate enough with distance. When I added convergence everything was right. But I still have to think about what it is happening geometrically. Furthermore, I still haven't tried the zero convergence technique described at the SDK documentation quoted above.

The main problem with fake 3D, and the reason why it needs some "training", is that it real life, optical focus of the lens in the eyes (what's the name in english? plz!) and convergence work synchronized but in the computer the focus is always on the display surface meanwhile the eyes change convergence. A beginner needs to use low separation and increase it as the brain gets used to this unnatural discrepancy.

[digitalmellon]

Yes, the eyes converge for near objects and look straight, or almost, for far objects, but the objects must be at the right spot for this to be possible. I have to think about it throughly, too. I found out the need for convergence when I modified a direct3D sample to add stereo. With separation there was stereo effect but when I walk away from the objects they didn't look as being far because the left and right images didn't separate enough with distance. When I added convergence everything was right. But I still have to think about what it is happening geometrically. Furthermore, I still haven't tried the zero convergence technique described at the SDK documentation quoted above.

In the last week i have tried to set convergance at 0 (using "Dragon Utility" from this site), but what i get was 2 identical separated images. As i set separation high, the strait lines did not follow to meet at vanishing point as i expected.

Don't know what to think right now. This is all so awkward.
I should continue my research. :]

The main problem with fake 3D, and the reason why it needs some "training", is that it real life, optical focus of the lens in the eyes (what's the name in english? plz!) and convergence work synchronized but in the computer ...

I'm aware of that, but it's not such a big deal though and we can't do much about.


"lens in the eyes" seems to be just lens.
http://en.wikipedia.org/wiki/Image:Sche ... eye_en.svg

[crim3]

"lens in the eyes" seems to be just lens.
http://en.wikipedia.org/wiki/Image:Sche ... eye_en.svg

Ah, ok. In spanish has a name, so I thought it was the same in english.

[shonofear]

so those videos still work....?
cause i cant get em to work with either IE 6, FF3, even the new google chrome browser.
I am using XP SP2 and WMP 10, maybe i need WMP 11...?

but i think that quick time is more reliable, personal thoughts though


Other than that good read Neil.

Cheers

[Neil]

The videos still work - just tested them.

Upgrade Explore to 7.0, and Firefox needs the Windows Media Player plugin.

I can look into the quicktime option though.

Regards,
Neil

[LukePC1]

This might help you:

I get the problem with the videos too on Firefox. It shows a black square in which it's written "(no video)" (without the quotes) and there's no play controls below the black square.
The videos work fine on Internet Explorer 7.

The content is good. I think it's well explained.

EDIT : I got the videos working.
I found this : http://kb.mozillazine.org/Windows_Media_Player
This lead me to : http://port25.technet.com/archive/2007/ ... refox.aspx
Which gives a plugin for videos on this page : http://port25.technet.com/pages/windows ... nload.aspx
I installed this plugin and now the videos work.

Or if you have a black box you might right click in FF to get the menue and then hit start/play.

Good Luck

[phil]

[EDIT: Sorry, I didn't realise 'convergence' had several meanings in 3D (how confusing)]

Hi,
It's good to see the maths behind stereo graphics being discussed here - I actually think this is really important for the success of S-3D.

D3DXMatrixPerspectiveOffCenterLH is indeed the tool for the job, if you want the effect to be true-to-life. I only found this out recently, by drawing some diagrams and seeing that convergence [EDIT: toe-in] makes no sense. Convergence [EDIT: Toe-in] just seems to be a lazy approximation of the asymmetric view frusta described above.
When you use the zero-convergence [EDIT: zero-toe-in] approach with asymmetric frusta, your stereo becomes deep, weighty and convincing - with natural pop-out when objects are closer than the screen plane. In other words it becomes accurate.

If it helps anyone, the source code that comes with my game (thread) shows D3DXMatrixPerspectiveOffCenterLH being used to create a geometrically-correct stereo pair for viewing on a single screen (there's a separate option for HMD).[EDIT 2: As CarlKenner pointed out, stereo for HMDs should be generated as if it's for a big virtual screen.] The gist is that you render the scene using 2 parallel cameras but then shift each eye's image across the screen so it's centred in front of the eye. This is basically how stereo should be done - not using any convergence [EDIT: toe-in].
The stereo parameters therefore become:
- interpupillary distance (the physical distance between the viewer's eyes)
- and screen width.
- [EDIT: and a "focus depth", which can go by many names...]

It doesn't quite end there though. For a start, experts say that the interpupillary distance should be artificially reduced (a lot) when stereo images are displayed on a monitor. For various reasons, many viewers will find it uncomfortable and confusing to look through the monitor to the objects' natural depths. Personally I love it that way.

But to summarise:
If you want stereo that makes some geometric sense, to create natural-looking depth effect on a monitor/projector, don't use convergence [EDIT: toe-in].
Use parallel cameras and offset projection matrices (asymmetric frusta).

Cheers,
phil

[Freke1]

Your're probably right (don't understand everything You write), f.ex. in 3D videos You natively have 2 slightly different views and just have to adjust separation.

In games however You only get 2 slightly different views of the gameworld by using some convergence.
2 separated identical images is not 3D, just 2D at a different distance than the distance to Your monitor.
You talk about native 3D games I think?

[phil]

[EDIT: Sorry, I didn't realise 'convergence' had several meanings in 3D (how confusing)]

Hi Freke1,
Games are the same as videos in that you use two slightly different viewpoints (I call them cameras) to render the scene.

To make the effect identical to what you'd see in real life (disregarding FOV), the game would do this:
- separate the two cameras by the distance between your eyes;
- keep the cameras parallel - no convergence [EDIT: toe-in];
- shift the two images apart on the screen by the distance between your eyes. This is done by offsetting the projection matrix.

In this way:
- when looking at an extremely distant object, your eyes are parallel (the object appears miles away);
- when looking at an object at screen depth, your eyes converge at screen depth (the object appears at screen depth);
- when looking at an object closer than the screen depth, your eyes converge closer than the screen (the object pops out).

But like I say, this kind of full-strength stereo isn't comfortable/desirable for everyone, especially when the viewing distance (from the screen) is small.
It is however a good starting point that we can adjust from. I think it's better to start to with an accurate model of how our eyesight works, and adjust from there, than to start with something completely made-up (converged cameras [EDIT: ie. toed-in cameras]).

As always, I'm thinking of native stereo games here, but the same principles should equally apply to stereo drivers. I gather they use convergence instead, which I find a bit worrying. Maybe they have some good reason for it that I don't know about Or could it actually just be lazy design?[EDIT: Apparently they use asymmetric frustum (see thread).] I can certainly tell you when I made my first stereo mode for a game, I converged the cameras, because it looked approximately right and I knew almost nothing about projection matrices back then.

[Freke1]

I think we use sonvergence the way we naturally see. Because when shooting 3D videos with parallel cameras You can't shoot close up objects (closer than about 1m). At least it's uncomfortably to watch it So I'm guessing there's some sort of "convergence" because in RealLife it's not uncomfortable. Just my thoughts though anyway I too am a big fan of when the monitor glass totally "disappears".

Another mystery to me is the amazing good 3D there is in a picture like this:
http://www.mtbs3d.com/gallery/albums/mt ... OutMix.jpg
It "shouldn't" look that darn good really. But thank god it does

If You use Your "parallel eyes" method does the close objects in the gameworld look ok?
Cheers.

[yuriythebest]

Another mystery to me is the amazing good 3D there is in a picture like this:
http://www.mtbs3d.com/gallery/albums/mt ... OutMix.jpg
It "shouldn't" look that darn good really. But thank god it does
Cheers.

yeah some anaglyph images look very good

[The_Doctor]

whoa! That's crazy

[phil]

Freke1 - Hi again. When you make 3D videos, do you shift the left & right images apart slightly after filming?
I think if you were using parallel cameras and displaying the results on a monitor, that's what you would need to do to make it look natural and comfortable. Otherwise, like you say, close objects would appear at the wrong depth and be too close to look at.

Using the parallel cameras method, objects closer than the screen do still pop out - your eyes have to converge closer than the screen to look at them (as if they were really poking out of the monitor). The effect can be altered by varying (1) the separation of the cameras and (2) the separation of the images on the screen, but those are things I've yet to experiment with. In my game I've only set it up to exactly mimic how you'd see the scene in real life.

The method I'm describing allows a full range of depth to be displayed:
from infinitely far away (eyes parallel),
to screen depth (eyes converging on the monitor screen),
to right in front of your nose (eyes converging very close).
And because the depths are true-to-life (dependent on the parameters being set accurately), it can give you an instinctive awareness of how big and how distant/close the objects are. The result is that you feel more like you're actually there - you can feel the 'weight' of the scene.

It's hard to explain, but if it sounds interesting I suggest you give it a try! Also, v21-onwards of my game uses this method so you can see it in action - I found it made a huge difference. But I should mention that game cameras are slightly different from real cameras in the way they project the scene (they don't fish-eye in the same way).

yuriythebest1 - Cool 3D fur in that photo!! Do you have a higher quality version, or the raw stereo pair?

Cheers,
phil

PS - Sorry for being a bit bold earlier, and for knocking the thread a little off-topic

[Freke1]

Hi - yes I adjust it (also during playback).

[soundstorm]

Hi Neil

Unfortunately, the videos are down in the 3D settings guide:
http://www.mtbs3d.com/index.php?option= ... &Itemid=98" onclick="window.open(this.href);return false;

Is there a possibility to up them again? The guide is very good, but the videos help a lot.

Greetings

[Neil]

Fixed!

http://www.mtbs3d.com/index.php?option= ... &Itemid=98" onclick="window.open(this.href);return false;

Our CMS was adjusting the scripts for some reason. Thanks for pointing this out.

Regards,
Neil