3D Gaming*

[Silent_Buddha]

3DS really does live up to all the hype by the way. And in a sense it's a mini-preview of how 3D should be: no degradation of brightness, no glasses required, etc etc. The only thing that isn't so good is that the screen itself is very small. Would've liked it to have been PSP-sized at least.

Yeah, 3DS is the only 3D device I've seen in the past decade that I've felt was actually compelling from a consumer POV. Not only does it work well from virtually all reports so far, but it also won't cost your average family an arm and a leg. Have tried and experimented with other systems to varying degrees of comfort/useability.

I'm going to get one as soon as I can, I'm already expecting there's going to be shortages of it.

Regards,
SB

 

[patsu]

Yes, it will help drive living room 3D, glasses or no glasses. Once you get used to the 3D perspective, you'll notice the missing volume in existing 2D games.

Just a little off topic note... I told you all that 3D is going to be big
If Sony can gather enough customer interest, then this feature is going to be a standard on the next gen consoles. I can see even the next Wii introducing full support, especially as it can significantly enhance the motion control based experience...

Phhhptt.. I thought so too ! -- especially after I saw Plasma 3DTV.

I'd like to point out that Crysis' method is not merely 3D. It's 3D + HD. If KZ3 can do "proper" 3D without dropping resolution, then yes it'd be an apple-to-apple comparison visually.

But a (subHD + "proper" 3D) vs (HD + "faked" 3D) is an open issue right ? You don't even need to bring up SD fatique issue since I don't think it's true.

People have headacle and fatique because of inconsistency in what the eyes see and what the brain perceives, not because the res is lower. I have read that 3D will raise the perceived res of the image anyway (Doesn't this means HD will become even more fine ?).

EDIT: _phil_ mentioned 2 techniques. Are they the same thing ? Depth-based approach and Copy/make a second image approach.

 

[Laa-Yosh]

Er.. it's both, take one image + depth and create the second image from this data. In fact it should be create two new images from one.

Basically you offset the various pixels to the left/right depending on their distance from the screen's perceived image plane.
If you want to object to appear to be more distant, you need to move it to the left for the left eye and to the right for the right eye. Crysis only does this because the further you get from the viewer, the less you have to move something.
And if you want the object to appear in front of the screen, closer, then you have to move in the opposite direction, and by a lot.

Here's a quick illustration:

The amount of the offset you use will control the depth of the 3D effect.

 

[grandmaster]

But a (subHD + "proper" 3D) vs (HD + "faked" 3D) is an open issue right ? You don't even need to bring up SD fatique issue since I don't think it's true.

Is there any confirmation at all that Crysis 2 in 3D isn't sub-HD?

Remember, the demo is on Xbox 360, which cannot support 1280x1470. Maybe it's running at 720p60 with alternating frames per eye? I'm not even sure that is supported by 3DTVs, but short of CryTek using a 1080p framebuffer, it would need to be sub-HD if alternating frames are not supported.

 

[draconian]

Er.. it's both, take one image + depth and create the second image from this data. In fact it should be create two new images from one.

Basically you offset the various pixels to the left/right depending on their distance from the screen's perceived image plane.
If you want to object to appear to be more distant, you need to move it to the left for the left eye and to the right for the right eye. Crysis only does this because the further you get from the viewer, the less you have to move something.
And if you want the object to appear in front of the screen, closer, then you have to move in the opposite direction, and by a lot.

Here's a quick illustration:

The amount of the offset you use will control the depth of the 3D effect.

Laa-Yosh, nice illustration. So, this is what Crysis 2 does?

I'm guessing killzone 3d renders both the left eye perspective and the right eye perspective simultaneously. 2 full renders. Thus, this is why they take a strong hit on resolution and frame rate.

crysis 2 otoh saves some computations? Pixels and geometry that is viewable by both eyes are rendered only once? Then pixels/geometry that are unique to either eye are rendered separately?

My guess/understanding.....
crysis 2's approximation 3d rendering = if pixels/geometry is viewable by both eyes, put these into both their respective left eye and right eye buffers, and do displacement(shifting the pixels left or right) for depth then render the pixels/geometry unique to the left eye in the left eye buffer. Left eye is done.

at this point the crysis 2 rendering requirements are virtually the same as a single perspective render. Now they just need to render the pixels/geometry unique to the right eye, and they are done?

If crysis 2's 3d is perfect and not an approximation, then I guess there must be someway to save computation when generating the common pixels/geometry for both eyes....

 

[_phil_]

Now they just need to render the pixels/geometry unique to the right eye, and they are done?

The depth buffer will give you a single POW.The pixels (you don't even consider this problem at the geometry level) covered or not by each eyes are taken from this POW and copied from this buffer in the direction of the other eye to fill the unknown parts of the image.(in fact , they might have to rebuild both eyes from that average position )
It's that or let a hole in the picture.There is no way to reconstruct the real data from an other camera position.

 

[Laa-Yosh]

I'm guessing killzone 3d renders both the left eye perspective and the right eye perspective simultaneously. 2 full renders. Thus, this is why they take a strong hit on resolution and frame rate.

Yes, that is most likely correct.
It's also the Right Thing to do as far as stereo 3D.

My guess/understanding.....

Unfortunately you're probably wrong with your idea.

What Crysis 2 seems to be doing is to render a single frame with no stereoscopy or anything extra at all, just your every day rendering. It also keeps the Z-buffer. Hence the lack of performance penalty.

Now for each pixel you have a Z value, which is the depth information - how far it is from the screen's plane. They also have to calculate a relative distance for the eyes' viewpoints (sort of an ideal viewing position, ie. X feet away from an N inch screen).

Then they begin a simple 2D image processing pass, moving pixels to the left/right to create the left/right eye's images. Using the Z-value and the above mentioned viewing distance, it takes very simple math to calculate how much you have to move the current pixel.

As phil has mentioned, the trouble here is that you don't have any image information for the holes that you create in the image when you move these pixels around. The enemy has covered up the wheel of the car, now he's moved and you'd need to see that wheel, but there's nothing to work from.

This is the 'magic' part, coming up with something to patch these empty parts of the final image. Having no extra information but the base image, they probably have to resort to stuff like what Photoshop's "healing brush" does, or probably something even more simple.
When you convert a movie from 2D to 3D like this (Alice from Disney) you have all the time in the world and can take some operator guy to paint something into these empty areas. There's even some significant research in this field, although using an algorithm to fill up holes like this in a sequence of images requires a high level of consistency and determinism.

In the end, it will never be as good as actually rendering from two points of view.

 

[patsu]

Is there any confirmation at all that Crysis 2 in 3D isn't sub-HD?

Oh so true. You're from DF. You tell me !

I don't really want to trust his words (without some proofs) after reading his interview above. ;-)

That said, my question is still valid. If someone does a HD + depth-based 3D game, it is an open issue whether it looks better/worse than a subDH + "proper" 3D one.

 

[Laa-Yosh]

There's always going to be a performance difference between the 2D and the stereo versions, and the question is what type of balance the developers would aim for. It's a simple question of math, trying to do more with the same resources will require a drop in quality/quantity.
Ideally, the next gen consoles should have enough raw power to run a stereo version at HD res, and use the extra performance to enhance the 2D version's image quality only.

30fps / 60 fps is not really a valid option IMHO because the game's internal refresh rate would have to be sped up too, and rendering only one view doesn't free up enough CPU power to do that.

720p stereo / 1080p 2D is a valid approach, twice as many pixels requiring more shader unit cycles (assuming the new consoles all have unified vertex/pixel pipelines) and twice the video memory.

Another option is to up the antialiasing quality, even with MLAA I think there'd be room for 2x supersampling.

There's also the chance to go sub-HD and utilize significant vertical scaling for the stereo version, rendering at like 1024x720 (with MLAA or something), in which case there'd only be ~50-60% of extra performance to spend on the 2D version. 1.5 oversampling and downscaling would remain to be an option... or just scaling back some parameters in shaders, shadows, using full-res particle buffers instead of 1/4 ones, and so on.

 

[Laa-Yosh]

Another thing to consider is the debate between 16 bit and 32 bit color depths that we had about a decade ago (got reminded by reinstalling Deus Ex 1 yesterday ).

Back then, Voodoo3 owners were saying that 32 bit is too expensive, requiring twice as much memory bandwith and performance. TNT2 owners on the other hand thought it to be the Right Thing, helping to get rid of dithering and banding artifacts, and believed that eventually no game will use 16 bit.
Nowadays, we don't even argue about wether HDR / floating point color depth is a good thing, but instead we all agree that every game should try to implement it somehow.


So, I firmly believe that eventually every Sony 1st party AA game will support 3D out of the box and it'll pretty much become the standard for them. Many 3rd party games will also follow to stay competitive; Xbox360 is another question, but it's a feature they probably can't ignore for too long.

Will it be enough to make 3D TVs a commodity is another question, but for the next few years and especially the launch of the next hardware generation, I expect some serious push.
A big question is if the TV manufacturers can develop some new, higher quality implementation of stereoscopy; there's a lot of promising research, but without a clear winner, there's no chance to get new technology to the market within the next 2-3 years. And of course prices have to come down to a more reasonable level, too.

 

[MfA]

Prices can't really stay high ... the changes for shutter glasses based 3D plasma screens especially are almost negligible.

 

[Laa-Yosh]

And yet it costs twice as much as any other regular 50" plasma

 

[MfA]

It takes a while for competition to drive margins down (assuming they aren't colluding, which in this industry is not always the right assumption). LG hasn't introduced it's 3D plasma's yet, so it's just Samsung and Panasonic.

They might try to keep prices artificially high to keep the higher margin 3D LCD tvs competitive ... of course only one has to break ranks.

 

[matthias]

In the end, it will never be as good as actually rendering from two points of view.

Of course, but the brute force method of standard 3d rendering is boring. Isn't realtime rendering about who is faking best?

So what about that. Alternate rendering from left and right eye viewpoints and fill the blanks with the information from the previous frame:

1. render frame from left eye viewpoint (correct)
2. displace right eye from left eye frame + z-buffer (as already explained)
3. fill blanks with pixel information from previous right eye correct rendering
4. change viewpoint to the other eye
5. repeat with 1.

PS: I'm new, please don't hurt me.

 

[Graham]

Of course, but the brute force method of standard 3d rendering is boring. Isn't realtime rendering about who is faking best?

So what about that. Alternate rendering from left and right eye viewpoints and fill the blanks with the information from the previous frame:

1. render frame from left eye viewpoint (correct)
2. displace right eye from left eye frame + z-buffer (as already explained)
3. fill blanks with pixel information from previous right eye correct rendering
4. change viewpoint to the other eye
5. repeat with 1.

PS: I'm new, please don't hurt me.

I just posted exactly this idea.
I was quite happy, as it seemed to be a clear logical answer to the problem.
And then I read your post, right above my own.

Sigh. I'm getting slow.



Welcome.



....

Basically, render left,right,left,right...

When generating the alternative eye view, use a pixel offset based on depth - and use a reprojection of the previous frame to fill in the gaps.
Then do some funky stuff for transparencies.

 

[Laa-Yosh]

Again, these suggestions sound like "render two alternating and differently calculated 16-bit images instead of fully 32 bit images". The industry is just going to have to accept the performance penalty if 3D manages to gather enough momentum, just as it accepted 3D acceleration, 32 bit color, and so on. Quake 3 was id's most returned game because it refused to run without a proper graphics card; but Doom 3 never had this problem and managed to become their best selling game to date. KZ3 might not look that good in 3D, but by the time we get to KZ4 people may not care about the 2D resolution at all.

Sure, stereo 3D is not guaranteed to take off - there's a lot of talk about potential deal breakers from health hazards to overpriced TV sets with underdeveloped technology. But it's a relatively trivial feature to add to games (even if some rendering techniques don't work as well in 3D) especially compared to movies where you need a proper camera and a lot of additional post effects work to get it right.
And if everybody does it, no game will be at a disadvantage for sacrificing some of the 2D performance. Then, anyone trying to gain a 2D image quality advantage with inferior stereo 3D would probably be roasted, just as Clash of the Titans has received universal criticism for its fake 3D.

It all depends on the combined effort of hardware and content providers, and since they really need some new feature to maintain customer demand, we can trust them to try really hard.

 

[BadTB25]

I wonder if Crytek's solution lends itself better to making older titles 3D. It would be pretty cool if they were able to make some of the older shooters 3D-like with minimal hassle--I'm looking at Halo 3, Gears 2, Resistance, etc here.

I believe Batman AA came out with a game of the year edition that was supposed to be 3D. Without having to rework the entire game/engine, how was Rocksteady able to accomplish 3D?

Edit: I will not buy into 3D until they are able to do it well without the glasses and limited angle view.

 

[draconian]

Laayosh, thanks for that explanation.

ok, so are we absolutely certain that crytek doesn't have some magic/secret algorithm? Anyone have press conference photos of that EA event?

Grandmaster, you're the DFguy, use your weight/pull and ask crytek for the screen shots of their 3d. Say people are doubting their 3d technology

And since it's for their benefit, they should provide screen shots to prove their 3d is real 3d. If they don't provide them, then I guess we still have an answer as to their 3d tech.

edit: ok, I found some camera captures from that 3d event.
LOL, let's see if we can even see anything...
here is the main link of someone's photo album
http://smg.photobucket.com/albums/v487/spdk1/E3 2010 EA Press Conference/?start=20

 

[Silent_Buddha]

Again, these suggestions sound like "render two alternating and differently calculated 16-bit images instead of fully 32 bit images". The industry is just going to have to accept the performance penalty if 3D manages to gather enough momentum, just as it accepted 3D acceleration, 32 bit color, and so on. Quake 3 was id's most returned game because it refused to run without a proper graphics card; but Doom 3 never had this problem and managed to become their best selling game to date. KZ3 might not look that good in 3D, but by the time we get to KZ4 people may not care about the 2D resolution at all.

Using that same analogy we're still looking at between 2-5 years until it's practical.

Similar to your analogy it's going to require hardware that is able to do full speed full resolution 3D before it's accepted by the masses. In console terms that means possibly next gen hardware. In addition TV manufacturers must cooperate by reducing prices as quickly as possible to tempt people to replace their TVs.

In the current global economic climate asking people to possibly replace a HDTV that is less than 3-4 years old just to play games and watch some movies is asking a bit much, IMO.

Sony is in a tough position. If 3D TV's take off they stand to gain a lot as they could sell them for a premium. On the other hand for 3D TV's to take off they have to be cheap enough to convince people in large masses to ditch their recently bought HDTV's and spend yet another large chunk of cash for a new TV. Which then removes the positive effects on company cash flow via large margins.

It "may" be as you said that eventually (the 16 bit to 32 bit evolution took many years from the first introduction of a 32 bit consumer 3d card) 3D will be standard rather than a gimmick. I remain unconvinced until technology improves significantly.

Just as most people remained with 16 bit, a lesser amount with 22-bit dithering, and even fewer actually using 32 bit when gaming. I feel that's going to be similar to what we have for the next few years. Except 3D will be in even smaller percentages than 22-bit dithering and 32-bit as the hardware (consoles + displays) will be a far smaller percentage than 32-bit capable 3D cards versus 16-bit capable 3D cards.

Enthusiasts and early adopters will probably rave about it. But your average consumer and console owner, probably won't be too enamoured, especially when they start looking at the costs involved for either fake 3D or sub-hd (with possibly lower quality or missing effects) real 3D.

Regards,
SB

 

[Shifty Geezer]

ok, so are we absolutely certain that crytek doesn't have some magic/secret algorithm?

We can be very, very confident. There's no way to automagically reconstruct missing data, despite what American crime TV dramas suggest!

edit: ok, I found some camera captures from that 3d event.

Ideally we want a view that has a deep object just off centre, or an object very close to the eyes, which these don't really show. Perhaps carefully selected to showcase the 3D in its best light. Still, looking at the hands of that figure it's apparent that it's the same image displaced left+right. Everything at this point indicates Crysis 2 is using the 2D displacement tech, similar to TVs creating 3D from 2D only with the added advantage of a depth buffer to get accurate depth info.