Meant to be Seen

Hi,

I think it would be good to gather everyone's opinions on what options games should have to control the 3D.

What exactly should each of the parameters do? What should they be called, and what should their limits be?

Thanks
phil

Yes, I think this is very important. There should be some common standards and naming conventions. Here is what I think:

There should be 2 values: Separation, Convergence

Separation: This controls the horizontal spacing between the 2 virtual cameras. At zero, they are at the same point in space. A value of 100% would be the recommend separation (chosen by the game developers, as each game may have specific requirements). Value is a floating point number. There is no maximum value.

Convergence: This controls the focus of the camera. At zero, both cameras are facing straight forward. 100% would make both camera face toward other such that each eye has zero overlap, and -100% would be both facing away from each other with no overlap. Value is a floating point number. This can also be set at "auto" which would allow the game engine to choose what to focus on. Care must be taken that the 3D engine renders with an asymmetric frustum, to avoid keystoning or other distortions.

The above controls are hidden within an "advanced" sub-screen of the 3D menu. On the main 3D options there is only a "3D depth" slider, which goes from 0% to 100%, integer values only. 0% is naturally 2D, 50% is the recommended default setting and 100% would be the extreme setting. The percentages really translate into separation/convergence settings that the developer defines. If you go to the advanced menu it will start out with actual values represented by the depth setting. This way you can find a good general setting and then tweak it if necessary.

I'm not sure if this is how the current 3D drivers work, it seems each driver has different options, which do different things. And the console developers all have come up with their own system. I am not sure how much common ground there is (or if my suggestion would even work). But its something to think about.

Just wondering: Would besides separation+convergence a horisontal parallax control be useful?

I originally was thinking that, but I believe convergence should be able to serve that function.

Beside these separation and convergence settings, I'd like to see another option added related to the real position of the observer eyes. That would allow a physically correct rendering (ala Johnny Lee), just like what is used with head-tracking devices.

I'd like to be able to specify the screen parameters (aspect ratio and diagonal in inches), viewing parameters (distance and height from the center of the screen in cm/inches) and real eye separation in cm/inches.

These values are already calculated to create the two cameras for S3D rendering, so I guess it shouldn't be that hard to provide such settings.

And I quite prefer to enter meaningfull physical values instead of using percentage values when I don't even know what they physically correspond to. That would end all this annoying trial and error dance that I'm obliged to do to obtain something satisfactory.

Well there you go Phil. Its actually quite a nice solid list. And very do able. I would also, add my own personal requests.

1) In first person shooters I'd like to be able to turn on and off all gun animations and hud info.
2) For racing games I'd like to use in the car view point and turn all car interior animations on and off.
3) I'd like to toggle on or off all hud info in all games.

With the ability to turn off and remove these near screen objects one will be able to crank the 3D up into the exagerated realm. Though its not real practical to play for long periods of time with super high 3D settings. It is cool to experiment around and enjoy the resulting crazy 3D.

cheers everyone

Yes, I agree that there should be options to remove clutter from the screen. This means hiding the gun, hiding the crosshairs, hiding the HUD, etc. I find this is one of the main reasons UT2004 is my main title for testing 3D. Simply because I can turn all those things off and crank the stereo up to the maximum value.

While being able to turn on and off different things is important, I think that it's much more important to be able to receive correct depth positioning of the hud and weapons in a game, and being able to set the convergence and separation of the weapon independently. We all know that the weapon is used to aim, even when there is no crosshair, but I can't imagine it to be so high in the 3d space, unless the player is constantly walking slowly or crouching. If he is running with a weapon in this position he'd be poking his own eye out. If anyone remembers the fix for Doom 3 and doubled weapon in the old Nvidia days ( the character got pulled back, making the gun's position lower and farther from the camera) this is what we need plus a centered crosshair. Now when the character crouches, or walks slowly the weapon should be raised automatically. As for racing games, if one plays in the cockpit, then no hud is needed, but when using a bumper cam an in the screen hud wouldn't hurt, especially if there is a 3d holographic map

Hi, thanks for all the feedback.

cybereality, I'd like to offer a couple of opinions that I think will straighten-out the ideas that you've given there.

The gist that I'm getting is that the player should be able to control not only the strength of the stereo but also the convergence plane (so as to choose the scale and closeness of the scene according to preference).

So at first it seems logical to have two parameters exactly as you've described: one controlling the inward shear of the camera frusta (ie. projection offset) and one controlling the camera spacing.

IT'S A TRAP!

Let me explain...
When you change the camera spacing, it moves the convergence plane.
When you change the projection offset, it also moves the convergence plane.
When you change the field of view, it also moves the convergence plane.

Since the convergence plane is the variable that we actually want to control, and since the required projection offset is the same regardless of camera spacing and FOV, the solution is to calculate the camera spacing dynamically based on these other variables, and never, ever control it directly.

If the programmers adding stereo to their engines would (please) take the time to understand this, and also use a properly-calculated projection offset, 3D in games would be a hit-and-hit affair. For example, this would have totally prevented the weakness and inconsistency in Gran Turismo (it's no coincidence that the strength is only correct in the cockpit view: that's the view with the closest convergence plane).

The equations are simple:

cameraSpacing = (2 * stereoStrength * interpupillaryDistance * tan(halfHorzFov) * zeroParallaxDistance) / screenWidth

projectionOffset = (+/-) stereoStrength * (interpupillaryDistance / 2 / screenWidth)

And as you can see there, the convergence parameter (zeroParallaxDistance) can be simply that, a distance. How that should be influenced by the player is another matter (I'll pick that up in my next post).

Anyway, sorry for the rant back there, just had to get that out of my system

Are people in agreement with my reasoning btw?


Thanks again for the feedback,
phil

Chiefwinston wrote: 3) I'd like to toggle on or off all hud info in all games.

Win!! You can do this on the PC however, works excellent, could be animated though.... Im turning it off with the press of my thumb on novint falcon FFB controller (mouse emulation). (playing Borderlands almost exlusively btw)
Ill read the other posts, cool topic.

Phil: then, the options should be interpupillar distance, screen size and distance to screen; plus stereo strength ranging from 0 to 1.

To me, as a simmer, that would be perfect as I always want realistic stereo. If the simulator could be feeded with those 3 parameters and calculate FOV, separation and convergence plane for me my life would be a lot easier. Instant full scale world wihout more hassle.

What I'm not sure is how this would fit to control stereo in other games where normaly you want a very different result. To begin with, game FOV use to be a lot greater than the real FOV of the screen, making characters and things appear the size of toys, which is fine for games, specially 3rd person ones.

Options to unclutter the view should not be needed in a game that is made with 3D in mind. Everything should be placed at a position and depth that feels right to the player. Despite of that, extra options are always welcome, specially for us PC users. We are used to have the control.

@phil: You bring up some interesting points but less user control is always a bad thing. I think the developers should expose ALL VARIABLES to the user inside an "advanced" menu. This means whatever parameters they are using the calculate the 3D: Stereo strength, IPD, distance to screen, screen size, convergence plane, etc. It should all be user modifiable with floating point numbers and no clamping. Otherwise you get into the situation you have on certain console titles, with pathetic depth and no way to fix it. I still think there should be a "simple" 3D menu for beginners, and just give them a 3D slider (like Nintendo has done) that will be easy to understand. But there should be a way to tweak this further. Everyone sees 3D slightly differently, and the same settings are not going to work for everyone.

cybereality wrote:I think the developers should expose ALL VARIABLES to the user inside an "advanced" menu. This means whatever parameters they are using the calculate the 3D: Stereo strength, IPD, distance to screen, screen size, convergence plane, etc. It should all be user modifiable with floating point numbers and no clamping.

I like your thinking, but camera spacing is not an input variable for my stereo calculations - it's the result of the other variables being controllable. It's the part of the mechanism that has to move freely to let the other parts go (and stay) where we want them.

If you take control of the camera separation, you lose control of the convergence plane (it becomes some arbitrary function of the FOV, screen size and IPD). From that point on, the stereo system is going to be broken, no matter how much time the developers spend trying to tame it and balance it by trial-and-error / hacks. The end result will likely be something weak and inconsistent, like GT5.

Does that explanation make sense? Are you sure you want to control the camera separation?

Now I see why the slightless change in camera parameters means having to readjust stereo settings with iz3d drivers. And that cross influence between stereo separation and convergence that make them quite difficult to set right.

Phil your post really opened my eyes. I always felt there was something a bit wrong with adjusting the convergance and separation like most games and drivers do but I couldn't place it.

My own suggestion is this: To be able to see the screen while I'm adjusting the 3D settings! The worst offender I've come accross is Gran Turismo, where you can't adjust the 3D while playing, and the settings are buried within the main game menus, so each time you adjust the settings you have to load the race again which takes forever.

Would also like the option to turn 3D on and off at any time during the game.

I'd just like to see a game that didn't screw at least one thing up in Stereo 3D, and made a half decent attempt at presenting a UI/HUD in stereo.

As for options, Auto convergence would be nice, other than that I'm reasonably happy with the way the NVidia Depth and Convergence settings work.
If the 3D Vision forums are anything to go by it'll be a while before the majority of users even understand that.