Meant to be Seen

Credit to 3DVision-Blog for spotting this!  We were very disappointed with our first run of Battlefield 3 in stereoscopic 3D because it had a depth-only experience, and seemed to be drab compared to what we grew to expect from its predecessor Battlefield Bad Company 2.  The good news is that there are indeed convergence controls hidden in the config file, but it's a pain to configure.

This utility provides a more convenient way of adjusting the convergence and getting the stereoscopic 3D experience you want.  Give it a try, and share your results in our forum.
Read full article...

Why is it that users need to resort to hacks just to adjust convergence. I almost feel like it was better before developers knew about 3D...

It's worse than that. Just tested it for myself, and DICE got their 3D terminology wrong. This "convergence" tweak is actually the adjustment for the game's separation level. Nothing to do with convergence at all. Bummer!

Regards,
Neil

Thats because Nvidia has it backwards, DICE was just following their lead. I have brought this up before and you didn't believe me.

Bizarre. Nvidia really calls Separation convergence?!?

Regards,
Neil

I hate being locked down in any area.

However, I have to say that the new games seem to really work pretty well with 3D Vision. I've been quite impressed that I can easily use the menus and UI in 3D mode, both in Skyrim and BF3, without any problems.

Not being able to adjust convergence however IS crap. Usually I like to jack it up just a bit so the guns or weapons stick out of the screen a little.

Neil wrote:Bizarre. Nvidia really calls Separation convergence?!?

What Nvidia calls "convergence" in fact controls (mostly) the separation between the virtual cameras. This is why you get more extreme 3D effects by adjusting it. And what they call "depth" changes *both* the separation of the cameras, and the convergence of the images (or you could say how much of the game is in front or behind the screen). So, in practice, both options adjust both values simultaneously. This is very confusing, and I am not even sure I fully understand how this works. But I understand enough to tweak it to my tastes.

No, convergence and separation terms are used as they should in NVIDIA terminology :
From : NVIDIA 3D Vision Automatic Best Practices Guide

Hi Guys,

Yes. Cybereality, from what I can tell, Nvidia presets a mixed separation and convergence setting for each game, and treats the emitter and "depth" control keys as separation of the virtual cameras. There isn't a dynamic formula between the separation and convergence as things are adjusted.

Regardless, Nvidia's language is still consistent. In their advanced settings, they refer to "CTRL-F5" and CTRL-F6" as convergence settings. The choice of calling separation "depth" was probably done to make it easier for gamers to understand the control.

Regards,
Neil

Have you guys used the Nvidia driver recently? I don't know why I am the only one that realizes this.
To illustrate my point, I have taken some screenshots.


This first image is with "depth" on the minimum, and with a low amount of "convergence": The second shot is with the maximum amount of "depth" and the same low amount of "convergence".
Take note that there is barely any difference in parallax between the beams, which indicates the virtual cameras have not moved much at all. This third image is with the minimum amount of "depth" but with a high amount of "convergence".
You can clearly see a visual difference in parallax between the beams, so much that on one eye they are overlapping, with in the other eye they are staggered. You can also see the gun is now visually separated between the views. So this demonstrates that the position of the virtual cameras has changed. In other words, the separation between the cameras changed even though I only adjusted convergence. This last image is with the maximum amount of "depth" and the same high level of "convergence". In this case there is a clear extreme separation effect, as you would expect with extreme settings. But it also seems to be very extreme, almost as if Nvidia has some algorithm with a multiplicative function. It is certainly not scaling linearly, as you would expect if the hot-keys were only changing two distance values. Or am I misunderstanding something, I am open to hearing other suggestions.

I've measured the parallax for 3 different points in the first 3 images :
- blue lamp in background : 2, 1 and 0 pixels ;
- front sight of the gun : 155, 115 and 51 pixels ;
- closest part of the gun : -13, -211 and -537 pixels.

You can see that there is much more parallax in the second image than in the first one, so the modification of the separation had an effect on the depth.

In the third screenshot the difference in parallax is due to the fact that the convergence plane is behind the eyes.

Illustration (note that the scale is not correct especially for the last image because of space constraints, but you get the idea) :

My third drawing is wrong, in fact the convergence plane should be farther than the whole scene with the rest unchanged. The end result should still be the same though.

Hi Guys,

I went about things the old fashioned way and asked Nvidia directly. The EA/DICE team are using their own stereo 3D descriptions, and this is not based on anything Nvidia motivated. I.E. Nvidia isn't to blame!

Regards,
Neil

I'd be interested to hear what Nvidia official stance on this is.

Everybody has their own "definition" of depth and convergence. I've called for some "official" definition a long time ago (if You remember).

Looking at the images by cyber, I think that they are correct.
In the second image, you can tell the separation has increased, so it has increased the depth. The actual convergence of the cameras shows that the gun is basically at screen depth.
In the third image, the gun is in front of the screen now, so the convergence has changed.
Part of the confusion, I think, is that they are using a correct frustum, rather then just changing the angle of the "camera" as I believe that some other solutions may use.