@mrfahrenheit94: Ok, I will take a look at Arma II tomorrow. Unfortunately I don't think I will be releasing anything until its ready. Mainly because this is going to be a commercial product, and also because some things are not finished yet. With the way things are currently set up it will be difficult to make a trial/demo version, so I may just have no demo, but some sort of money-back guarantee. None of this is really final yet, but that's what I am thinking.

Since you've got Mirror's Edge working, could a similar approach be used for other Unreal Engine games? I'm thinking Tribes Ascend, playing that as a pathfinder in the Rift would be insane (mouse emulation would probably be best off with it since it's so fast, but the Rift's high FOV and 3D would still be awesome).

Also Crysis Wars, so I can play MechWarrior: Living Legends

A version with full functionality that cuts out after five minutes shouldn't be too hard right, as a demo?

@rajveer: I figured other Unreal games would work (at least UT3 does mostly) but I attempted Bioshock and it has a lot of problems. So it still seems each game will need some custom work.

@Chriky: Yeah, I could probably do something like that. I just want to make sure I am spending most of my time adding support for games and not spending it with copy-protection/DRM. But I guess I have til around Nov/Dec to have things finished so I should be able to figure out something.

Shame to hear about Bioshock. One of my favorite games ever!
How much work do you think it'll be to get it working?
If it's not a rude question, what sort of price point are you aiming for once complete? Are you looking at an upfront cost + additional fee for each game? I just think that it could be good in terms of keeping it viable after release if you have some income for your after release efforts. It would be better for both you and customers if their initial investment can continue to be usable for many future games... assuming they don't all start including native support.

@android78: I want to have a price-point that is competitive with other similar products out there, so probably around $50 give or take. I was thinking that would cover all updates for a given release number (i.e 1.x numbered releases). And then when a 2.0 version came out, I could give existing users a discount to upgrade. But I think that is a while down the line, once I have support for a lot more titles.

And, yes, I do want to get Bioshock working so I will continue to investigate this.

Awesome! Probably a good way to go with the pricing.

With your driver, will it be possible to have HMD roll mapped to your hacked game camera roll, and pitch/yaw mapped to (wiimote acting like a) mouse ?

Yes, that would be possible. Not sure I can have Wiimote support by launch, but you could use my driver for just the roll and then another software (GlovePIE, FreePIE) to handle the Wiimote.

I've added a number of new display modes for the driver: anaglyph and all the major frame-compatible modes.

So what is supported now:

Side-by-Side w/ Warping (RIFT)
Side-by-Side
Over-Under
Horizontal Interleave
Vertical Interleave
Checkerboard
Anaglyph - Red/Cyan
Anaglyph - Red/Cyan - Grayscale
Anaglyph - Yellow/Blue
Anaglyph - Yellow/Blue - Grayscale
Anaglyph - Green/Magenta
Anaglyph - Green/Magenta - Grayscale

Been working on Bioshock this whole weekend. Got something about 80% working.



The main thing broken is smoke effects and stuff like that. They only render in one eye. Otherwise it looks OK.

Awesome work there, but it seems out of proportion - stretched vertically. Do you think you'll have to turn the smoke effects off, or will they be fixable?

Well the Bioshock image is from a SBS mode that would work on an HDTV, its supposed to be squished. This is not the Rift mode, I am just using it for testing.

I am hoping the smoke stuff can be fixed, will need to spend more time with it.

Sorry, all cool then. For some reason I thought you were doing these for the rift.

Please add support for Crysis/Crysis 2 ! (Crysis 2 may be easier since they already have builtin stereo code).

I'm not sure about crysis 2 being easier because they have inbuilt stereo code. If the PC version is like the xbox or ps3 version, then they are using some tricks to generate the second view that would be more restrictive then helpful. If the PC version does use this, then I think you'd probably have to throw it all out and use the full custom driver for the entire thing. At least, they should be rendering effects at the correct depth though, I guess.

Yes, Crysis 1+2 are on my list of games I want to get supported.

If only because when the inevitable question arises about the Rift: "But does it run Crysis", we can say "Yes!".

Since you've already added roll, have you considered adding head translation offsets? This would be distinct from moving the character's body position. It would essentially be a limited radius around the central character position (the hips) that you could adjust the camera position. So a 6DOF device could be used for subtle head motions and leaning or bending.

I think it has that brantlew, its how the stereoscopic effect works. In one of the early videos it shows the camera moving sideways away from the gun which stays at the player location.

Cyber, as long as we can install the driver after a hardware upgrade or os re-install and license it locally I'm all for that price point
The idiotic licensing scheme that DDD have is the only thing stopping me from getting their driver. If you buy something you should own it.

I think those were just his R&D tests to show that he could move the camera into the correct orientations to implement stereoscopic views. Now I want to know if he plans to put an interface that allows the entire stereo camera to be moved.

You can choose the stereo rendering display and algo (fast or dual) with console parameters.

Concerning the shifting of the camera rotation/position, except for roll and stereo I guess it's more complicated than that, since you have to hack the engine code itself for proper geometry drawing, not just the directx process.
...Anyway if you're able to do that it would be cool to have support for the most common protocols like VRPN, TrackIR, FreeTrack...

@brantlew: Yes, I can do translation (its essentially doing this for the stereo render). Not sure that will make it into the initial version, but I will see what I can do. Altering the camera will not take long, but providing an interface for using tracking devices could take time. Would like to do something generic that would support VRPN and stuff like that, but this will probably come later. Just trying to focus on what will be available on the Rift, which I believe is the Hillcrest Labs tracker.

@artox: I plan on being very open with licensing. So no extreme-DRM or anything (like locking it to a specific hardware configuration). If you buy the software, you can use it on any computers you own, no problem.

Great to hear, there's definitely a hunger for another driver. I have a few questions -
1.) how is the stereo (iz3d vs 3dvision), you've tried both drivers, so you probably know what I mean. People have been complaining that 3dvision offers less pop-out, but the weapon is rendered correctly and not split, which is the case with iz3d.
2.) performance-wise where does your driver stand
Thanks for the info in advance

Isn't it the other way around? Simple stereo projection option in the iz3d-driver takes care of these issues with some games while nvidia does not. You should have uncomfortable doubling of the gun while using the nvidiadriver (source games especially) when altering the convergence for a better 3d-experience. It's an issue with the games what i know about. The gun isn't rendered properly 3d-wise but when gaming in 2d it isn't disturbing. It's obviously not enough to make gamedevelopers look deeper into it.

@artox: To be honest, the stereo looks good, but its not as fully featured as the other drivers. But considering the Nvidia driver has been in development over a decade and I have just been working on this a few months. Whats working now decent is the camera separation, but I have some trouble with convergence with some games. So its decent enough to play, but could still use some work to be more flexible. I am also not doing anything special to help with rendering the gun, so if you crank the 3D up all the way then the gun will be totally doubled or off the screen.

The performance is also nothing special. So it takes around a 50% hit (maybe a little less). I can probably improve this at some point, but right now I figure people will have decent hardware, or can lower settings.

My main problem will basically all the previous drivers is that I believe they are not actually rendering correctly, and using kind of a cheat to make them work somewhat.
To explain further, they are using a symetrical view frustum, and to make the view appear 'more' correct, they angle the cameras inwards. To implement correctly, they should actually change the view frustum to make it asumetrical and have the cameras always pointing parallel.


I haven't done any real research into how the stereo drivers work, but is it possible to do this level of adjustment?

The following actually better explains what I'm talking about, and some implementation notes:
http://paulbourke.net/miscellaneous/stereographics/stereorender/

Yes, its possible to do this and is actually being done by existing drivers. Currently I don't have this setup yet, so I just keep the camera facing straight forward to avoid the toe-in distortion.

I don't agree with that, I hate simple stereo projection, it looks unrealistic and generally destroys the immersion for me. When playing with iz3d I prefer do hide the gun if possible and play with crosshair or without anything. With Nvidia this is not the situation, the gun looks to be on another plain and you can get much more separation if not convergence without splitting it. If I want to get the same amount of depth with iz3d I have to reduce the convergence severely or the gun is irreparably split.
The best solution would be automatic profile switching when aiming and showing less of the gun when not ( what was achieved via console editing of the fov in doom3 with the old drivers, the camera was pulled back). Another solution would be to only show the weapon when using iron sights and hide it at all other times, which should be combined with automatic profile changing. For me, this is the most realistic method, as it gives you better depth and convergence during normal play and it is pretty hard to run and aim precisely (weapon on shoulder) in real life.
Another thing I have noticed with Nvidia, opposed to iz3d is that after a certain point in the scene there is no convergence, only separation (as is in real life, after a certain distance you can't distinguish which object is closer), this leads to more ghosting and less pop-out of distant objects, but makes for greater depth. I guess this is achieved by calculating the distance between the closest and farthest object.
@android78
It would be interesting to see a comparison of both rendering methods. I think I'll give this paper a detailed read

p.s:
iz3d supports assymetric rendering, does this translate to better (correct) viewing, as is suggested in this paper, or is this someting else completely?

So I was having a lot of problems with Bioshock actually, more than just the fog. Gonna put it on hold for now and try some other titles. Anyway, I added a mono mode that has better compatibility. In theory, it should work with any DX9 game. What it does is handle the aspect ratio conversion and does the pre-warping. I could also do 2DOF tracking (mouse emulation) but its not in there yet. Can also alter the FOV, but its better to do this in game. So while its not stereo, it at least broadens compatibility to a lot more games.

Here is Bioshock in 2D mode with Rift output, in game option to 100 degree FOV.

I think 2D is an important mode as long as you are retaining the original frame-rate. The 2D mode doesn't require two-pass rendering does it?

The 2D mode is pretty cheap. It just needs 2 simple texture copies and a run through the pixel shader. I haven't benchmarked but I can't imagine this costing much.

Sounds good. Even though the Rift is low-res, a backtop system can still bog down with high quality settings and stereoscopic rendering so it will be great to have the option to go to 2D higher frame-rates.

Hey Cybereality, you sure are a hard worker man. I think your doing a wonderful thing and are very smart.

So since your working on a driver, I wanted to ask you and wondering if in therory that you can project your DIY autostereo project into your driver. Technically you don't need a sheet to block those lines. You could interfere the pattern to hold static black subsets on a 2d panel for left right view eh? I think so. Like to test could'nt you just load a pattern you have in desktop then transparent slide, block it over the source image?

Could definitaly make your driver more popular to those people with 2d panels also that you don't need to print them. And for compatabily just have a few differnt sub sets. Maybe I could help with some of the math. Let me know, thanks man.

Well the driver already supports vertical interleave. So it could be used with a parallax barrier. However, the barrier itself still needs to be physically printed and attached to the monitor. Its not possible to emulate this in software if that's what your asking about. That doesn't really make much sense.

Got DiRT2 almost working. Looks really good in first-person mode (no car) but the car is a little strange. Looks OK in third-person mode with low depth. Also, there are problems with the shadows (must set on very low). But its playable if you tweak it.

Yeah I thought the same thing and I am not for certain but with physically having a barrier sheet all your doing is blocking lines of pixels, one side the left eye and one side for the right eye. Those lines are pure black that totally block out whole lines like interleave except interleave does not show total black lines that are blocked out and are working together to switch between those left and right lines, I think. For some reason I though that with a parallax barrier it would just do static black hold without any interleave switching. Or is interleave just a static left right hold pattern too.

If you had the black lines barrier in software that mimicked the physical sheet using interleave and barrier cover it might work. Would there he a herz issue? I don't want to keep pusing this idea but it does seem to me like it would work.

Like would'nt turning off or making the lines on the actual pixels work as a barrier. Or have them switching along with interleave on black or off. By a slide feature if this does work is that it would make the situation a whole lot easier by being able to slide the software barrier to fit your screen pixel pitch right.

I don't really know but I just had to tell you.

Also I seen that you could put text into directx, so I was wondering if you could put a software parallax barrier like the text except black lines. Then run interleave but I guess they would have to be both switching at 60hz, unless they are static always on.

I think you are mis-understanding how parallax barriers work. What they do is block certain pixels from each eye. So your left eye will see (for example) only the odd columns and your right eye will see the even columns of pixels. Its a passive technology, so the sheet can be as simple as transparency paper with black vertical lines printed on it. In the case of professional parallax barriers (like the 3DS) they use a second LCD screen to show the barrier, but its still a static image. Not flicker, it has nothing to do with page-flipping. Even so, its a physical property of the monitor. Please do a google search of parallax barriers, or even watch my video on YouTube and I think it will make more sense.

Thanks for listing Cybereality, I'll check out some of those more specific topics on parallax barriers.

BTW did you get a copy of Diablo3? If not I can pick you up a copy or give you a cd key.

Sorry if I messed up your thinking patterns