Meant to be Seen

@tritosine: Thanks for the video, very cool.

Great news, If you skip to 14:45 of the video he says dual-render is still supported (maybe only on PC?), awesome stuff!

Also, he says the code to re-generate occluded pixels is only a few lines of code. I'd really be interested to see those lines...

Thanks for the video link tritosine, looks like it will be all from the screen depths inward. Still that will be cool for me. I'm not expecting the world for $60. Just some rocking gaming for 2 to 4 weeks. Then I'll begin the search for the next rocking game. This looks like one of many good games for 2011.

Cheers everyone

Cyberreality, have you ever created a 3d model. Then created a uv map. That let's you paint your models. It's cool. But it also gives hidden clues as to how the occluded areas. Are actually available at the game engine level. The occluded areas can be located on these maps. You wouldnt be able to do this with a camera photo. But with a game engine the info is there.

Cheers everyone

Chiefwinston wrote:Cyberreality, have you ever created a 3d model. Then created a uv map. That let's you paint your models. It's cool. But it also gives hidden clues as to how the occluded areas. Are actually available at the game engine level. The occluded areas can be located on these maps. You wouldnt be able to do this with a camera photo. But with a game engine the info is there.

Yes, I have done some 3d modeling and texturing before but it has nothing to do with this. You are mis-understanding how Cryteks method works. They are using what is called "screen-space re-projection", which is a pixel shader. When the pixel shader is run, it has no concept of polygons, UV maps, etc. All it has is the final rasterized 2D image and also the grayscale depth map. So all it is doing is re-positioning those pixels based on their depth, and then filling in the empty space with something (probably stetched/blurred pixels). It cannot access any of the information about the occluded space, as this would require a second render, and then you lose the performance benefit of using this technique. In that case it would make more sense just to do the dual render.

Having another look, there are definitely anomalies. It appears that the missing pixels are recreated using the pixels from the background being duplicated. It's hard to explain so I've tried to take some photos to show this -

The following is a full screen photo. You can see the worst of it on the sides of the gun:


The following is a close up of the side of the gun to show what I mean:


These images are taken when in scope mode, it's hardly noticeable when not holding the left trigger in.

Sorry cybereality, I know your firmly in the camp that 2D + depth can't work that well. I'm mearly trying to point to possibilities that occluded zones are not necessarily missing in a game engine world. I'm encouraged by those photo's. I'm in the camp that there is enough computer data to make fantastic 3D. That the brute force method is not the only way. The mathematical wizards are only just beginning to do their thing. A 2D + depth game engine that works great will be worth a fortune to some company. These wizards will have the next generation quake or unreal equivalant game engine.

cheers everyone

A game engine that works like this could concievably be worth $500 million to $1 billion.

cheers everyone

Thanks for the screen shots android78.

cheers everyone

Those screenshots looks exactly as when using the virtual 3d mode in tridef at longer distances. It's almost like a 2d surface where only the gun is in 3d (only looked crosseyed though). I'd like to know what happens if trying to turn up convergence a few steps. With tridef the image becomes really ugly when objects are reaching close to popout/popout (maybe provide other shots when convergence changed, if it's adjustable of course).
To be honest: A 3d-effect close to this can be accomplished by just turning one projectors (in a dual projector rig) thus shifting one image sideways to give more "depth". The image is still 2d even if experienced behind screendepth...

I think:
If this depth3d works so well, why not use it in 2D mode as well to simulate 60 fps when you only have 30.

Maybe it not works so well.

Likay wrote:Those screenshots looks exactly as when using the virtual 3d mode in tridef at longer distances. It's almost like a 2d surface where only the gun is in 3d (only looked crosseyed though). I'd like to know what happens if trying to turn up convergence a few steps. With tridef the image becomes really ugly when objects are reaching close to popout/popout (maybe provide other shots when convergence changed, if it's adjustable of course).
To be honest: A 3d-effect close to this can be accomplished by just turning one projectors (in a dual projector rig) thus shifting one image sideways to give more depth. The image is still 2d even if experienced behind screendepth...

The photo is taken when in scope mode. It seems that they actually reduce the depth of the scene when switching to this mode, so it is a little better then you see here. The reason I took the picture in scope mode is because this is where the anomaly is most obvious, along the sides of the gun.
Like I've said previously too, this is on a 58" screen too, so the 3D is reasonable at that size, but anything less the 50" would probably not be worth much at all.

I do believe that they are rendering the scene from a mid point between the eyes, then using a second pass pixel shader that takes the rendered image as a bump mapped surface, using the Z-map. For the pixels that are occluded from the initial rendering, they just copy the pixels beside. So if there are 10 pixels that were occluded for the right eye then they get the 10 pixels to the right of the occluded pixels and duplicate them into the occluded pixels. This means that you get similar colours and textures to the background and should be close to correct, the effect though is almost as if the close object has glass surrounding it. To be honest, I hardly noticed it when I first played the game though, so it can't be that bad.

I think that in future we will see some changes to the rendering procedures so as to properly render the missing pixels for these cases... I have some ideas myself as to how this could be achieved.

Chiefwinston wrote:Sorry cybereality, I know your firmly in the camp that 2D + depth can't work that well.

Well that is not exactly true. I was actually a huge supporter of 2D->3D conversion (which is mostly 2D+depth) and even spent some months writing a 2D+depth image conversion application (which I never finished). I abandoned the project solely because I could not figure out a suitable algorithm for re-creating these missing pixels caused by occluded areas. It is a very difficult problem that most people do not appreciate. And the easy ways to solve it do not look very good in my opinion. Technically there are things you can do, like Photoshop CS5's "content aware fill". However this is no where near being able to run in 16 milliseconds, which is what you need for real-time gaming. So in order to keep the performance benefit, you can to do cheap tricks like copying pixels, stretching them or blurring them. Its not very accurate, so then you also have to keep the 3D depth low to mask the problems. It looks like Crytek has come to an acceptable compromise, and I would like to see this for myself. But it is still limited compared to a true dual render. However, I am still watching this space. If I can learn something from Crytek, I would love to pick up development on my program. I would very much like to see it work.

Wow I thought the 3d done in Crysis 2 demo is one of the best examples of 3D gaming to date. Fist there is no ghosting noticable on my Panasonic VT25. In my opinion the 3d redering in Crysis 2 really shows how bad Black Ops 3d version is, with Black Ops downgraded 3d graphics, slowed framerate, and no 3d strength adjustment. All of which Crysis has non of these issues that i can see. The 3d used is all about depth, there is no pop-out of screen 3d. And it is the most comfortable 3d game played on lengthy periods with no eye strain.

Even James Cameron, the 3d man himself, said that he is impressed with the 3d from Crysis 2. And he claims the game's 3d is at Avatar's status. That means alot comming from the king of 3D himself. Take a look... http://www.gamesthirst.com/2011/01/14/j ... cysis-2-in.." onclick="window.open(this.href);return false;.

I'm not sure but I thought I gathered from Tritosines' video post that out of screen effects and seperation levels were more about their research into what works good for a mass audiance (not making them sick). And not neccesarily a limitation to there 2d+depth tech.

cheers everyone

Cybereality i don't think Crytech is going to give up the secret formula. There will be patents and intellectual property issues that will need to be secured before everyone learns the new trick. Besides there will be an entry fee for the deep down info.

cheers everyone

Chiefwinston wrote:Cybereality i don't think Crytech is going to give up the secret formula. There will be patents and intellectual property issues that will need to be secured before everyone learns the new trick. Besides there will be an entry fee for the deep down info.

cheers everyone

I don't think there's a big secret here. The occluded pixels are just filled with the adjacent pixels from the furthest layer. That doesn't mean there's no intellectual property involved though, so you may want to be careful if you're considering using this for a commercial product.

Chiefwinston wrote:Cybereality i don't think Crytech is going to give up the secret formula.

They don't have to give up their secrets. I can just look at the files in the PC version.

ahhaha hahhah.

cheers everyone

Chiefwinston wrote:ahhaha hahhah.

Huh

Well I'm hoping we get a good story mode game. Something that moves along cohesively. With good cinematic action. I'm more worried that it will play like a tech demo now. That's kinda the route that was taken with the unreal engine. I'm hoping we get something as engaging as Batman:AA or uncharted 2 story wise. There are a lot of mp fps. The weird lookin mechs and the nano suit work for me. Some of the animations didn't look like there was enough frames of animation for the speed of the game. All minor though but give it a tech demo feel. In any event we are still looking at one of the premier 3d games of 2011.

I think I'm more concerned with actual game play now.

Cheers everyone

It's gonna be cool. I just finished Crysis;warhead. It was cool.

There is a good bit about Crysis 2 3D here: http://www.eurogamer.net/articles/digit ... emo?page=2" onclick="window.open(this.href);return false;

I've just been testing the "unofficial beta" (leaked last week) and 3Dvision seems to be working ok. There are some problems, mainly with the gun position, probably because of the stupid FOV angel they have chosen, but hopefully this will be fixable with a 3D profile. I have not been able to compare it with the Crytek "Screen Space Re-projection" 3D because I can't get it working, it just seems to default to side-by-side mode, but anyone with a 3DTV should be able to test both systems, interlaced and dual-output modes are also supported. Hopefully we will able to use both types of 3D in the final game. The graphics look good, but nothing extraordinary considering its supposed to be a new engine (not like going from Unreal engine 2 to 3 or IDTech 3 to 4) but maybe the DX11 tessellation effects will improve things a bit. I have to say I think I prefer the original Crysis, it’s just my opinion but Crysis 2 feels too much like another Call of Duty clone...

no stop comparing to COD, crysis has a horizon that's much broader than ZOMBIES and boring year2year installments. Recognise that Cod is as generic as it gets , and not much about crysis2 is generic. You can forget about the default multiplayer once 32 player mods start appearing. Theres a full blown mecha mod already with jetpacks for CE2.

There are a lot of haters out there: http://www.metacafe.com/watch/5955316/h ... beta_leak/" onclick="window.open(this.href);return false;

Don't get me wrong, it's WAY better than COD, but I really liked the free-roaming design of Crysis 1 (the first half of it anyway), Warhead was a lot more linear, with more set-pieces, and Crysis 2 seems to be going the same way. Big set-pieces can be fun the first time but if I wanted to be stuck on "rails" I'd get on a rollar-coaster! I'm sure the modz will help fix some issues, only time will tell.

I sure as hell hope we get better AI than what was in crysis1 , and that better AI going to adapt to terrain and use cover unlike crysis1.

It has nothing to do with COD. Crysis2 gfx with smokin skycrapers looks like a masterpiece.

The clip was funny when it asked "Could all the people who play console games please leave" - the rest was just vulgar.

Regards,
Neil

yeh, these Downfall parodies have been done to death now, and there are better ones, I just thought it was funny someone took the time to do one about Crysis 2. He makes a few good points though. What's the point of having different hardware platforms if they all wind up running the exactly same games, the economics demand we will just get with the lowest common-denominator.

Ok, I have played the PC MP demo and it is, unfortunately, just what I feared. That is not to say it was bad, it just wasn't what I had hoped for.

First the good: the game itself was really nice. Playing in 2D, you get a silky-smooth frame-rate. This is not the un-optimized hack that was Crysis 1. This is a solid performer. And the graphics looked awesome. The gameplay itself was actually pretty tight, I can see myself getting into this game. They definitely went more in a Halo direction, but I think that will be popular. Turning on the 3D mode there was barely any drop in performance and it was still very playable with all the effects/shaders intact.

The bad: 3D mode is natively rendering but not natively output. So my Zalman interlaced monitor was not supported out-of-box. I had to enable the Nvidia 3D driver for the menu option to unlock. Meaning Zalman owners with AMD cards may be SOL. It is also not clear how they are detecting if you have an HDTV (to enable side-by-side mode), so I can't test this. But I would have liked something more like the Avatar game, where you could just choose how you wanted to output. But I could have lived with this if the 3D was good, which leads me to my next point.

The screen-space reprojection they are doing does not look as good as real dual-render 3D (as one would assume). The depth is really minimal, even at the maximum setting, although it does look more impressive than 2D mode. But its just nowhere close to what I was hoping for. While it did look OK, and was a comfortable experience, it just didn't add as much as it could to the game. It also sets a bad precedent for game developers to take this easy/lazy implementation instead of doing it the real way. In addition, it has some anomalies due to occluded areas (as I have been saying the whole time). Basically what it does is fill the invalid pixels with a clone of the adjacent space. So if you think of scanlines, if it finds 8 pixels that are blank (because something was moved) then it goes to the other side (from the moved object) and takes those next 8 pixels and copies them into the invalid area. This is actually not a horrible method, and is very computationally light. However it creates a strange "halo" effect around objects (like the gun in particular). This is most apparent with detailed patterns of textures with text or icons on them. Look at the following examples:
As you can see, this can look pretty bad in the extreme case. However it is not readily apparent when playing, so its not so bad. This is also the reason Crytek had to keep the depth/separation so minimal. Any higher and these anomalies would have killed the game. I was hoping they would have come up with something more advanced then this simple "clone" method, but I guess its still acceptable on a basic level.

Luckily the game is still supported by the IZ3D driver, not having anything to do with this screen-space reprojection business. I found the IZ3D mode to give much better depth and overall sense of 3D. However there were the usual glitches: shadows were offset in one eye, shaders/post-processing didn't work fully, and the gun was permanently blurred, among other things. Not to mention performance easily dropped in half. So while the depth was greater, it wasn't necessarily a better experience. I would still rather play with Crytek's "fake" 3D than with "real" 3D using the iz3D driver. But maybe some graphic setting tweaks could help alleviate some of these issues. Anyway, here are some cross-eye screen shots to compare. The Crytek/Nvidia method is on top, while the IZ3D is on the bottom. You be the judge.

well I'm betting this game rocks. fake 3d or not I'm ready to rock and roll with this one.

cheers everyone

I gotta tell ya. Playing the game with that blurry gun in your face all game will suck.

cheers everyone

cybereality wrote: The bad: 3D mode is natively rendering but not natively output. So my Zalman interlaced monitor was not supported out-of-box. I had to enable the Nvidia 3D driver for the menu option to unlock. Meaning Zalman owners with AMD cards may be SOL. It is also not clear how they are detecting if you have an HDTV (to enable side-by-side mode), so I can't test this. But I would have liked something more like the Avatar game, where you could just choose how you wanted to output. But I could have lived with this if the 3D was good, which leads me to my next point.

SADLY i can confirm this...

i got a zalman screen...
i got ATI graphics...
i got greyed out "3D OPTIONS"

SOL

Im on Zalman and Geforce and the 3D-Option is of course not greyed out. But after activation i only think

WHAT HAVE YOU DONE, CRYTEK ?!?

The in-game-3D-Fake is just a joke for anybody who played real 3D for some time. Its just not 3D, there are ~5 Layers of different depht, thats all. The menu und text is rendered in real 3D. WOW! The rest is not - no wonder this hits the performance only by ~3%!!
The pro: no ghosting etc. (of course, because there is nothing really 3d!!!).

You can activate the 3D-Vision-Driver, but the results are terrible: shadows, ghosting, flickering - all the way. The companies are complaining about copied games - here you have the answer why. The same happend weeks ago with Bulletstorm - problems ONLY !!!

CRYTEK, RELEASE A 3D-VISION-PATCH !!!