3d stereoscopy data redundancy

[serpretetsky]

So, if you've ever looked at a pair of stereoscopic images, you'll notice they usually look almost the same (very slight angle difference)

As far as i know, all stereoscoping rendering on current 3d video cards must render TWICE as many frames for a stereoscopic scene vs a regular scene (please correct me if im wrong).

But i'm wondering if that is really necessary. Since so much of the data on the second image is the same as the first, is there some possible method of extrapolating most of the second image from the first, and rendering signficantly less for the second scene, allowing the video card to save a lot of its rendering juice?

Couple of examples
1) sterescopy fails to produce depth data once the distance from the viewer is a certain factor greater then the distance between his eyes. The only reason you can tell one mountain is closer than another is because one of them is hazier than the other. Is there a way to make the video card not bother rerendering data that is a certain distance from the in-game camera?

2) most objects that are closer to the camera will have almost the same shape and textures on them, except rotated in one direction. Could the texture on that object simply be offset to one side and then that little sliver that appears on the other side be rendered, instead of the whole thing?

Is there any research into this?

 

[omeds]

TriDef has a mode called "virtual 3D" that extrapolates a 3D image from a single 2D frame, however it doesnt look anywhere near as good and often objects are not at correct depth.
I'm guessing its difficult to do, and to do it correctly may take some serious processing, in which case, might aswell just render the two correct frames in the first place.

I dont know of any "mixed" modes, of combining real 3d for certain objects, and extroplating the rest like in your suggestion, apart from Crysis 2, and the 3D effect is rather poor.

 

[serpretetsky]

Well, i'm not refering to attempting to extrapolate 3d data just from a single frame and nothing else. I believe this is what tv's to, and quite honestly, i have no idea how they even manage to do this since i can't understand how you get depth data from a single frame or even a stream of frames. I've heard they can attempt to use edges, motion blur, focus blur, etc etc, but in the end i still can't fathom how they do it. Maybe its unfathamability is part of the reason most people say it looks like crap, cause it really doesn't work. I don't know.


Here's what I'm saying:
The video card has two things to work with:
1) the single frame that has already been rendered, but is from the other angle
2) ALL of the depth data for the models being rendered.

 

[omeds]

Yeh I beleive Crysis 2 implementation is something along those lines, however the results are not very good.

 

[KingFatty]

I think the more you approximate, the faker it would look (i.e., instead of a smooth 3D depth to everything, it would look like a series of cardboard cutouts corresponding to various depth "steps").

As far as the rendered frame vs the depth data, maybe you can think of it as you have a 3D movable camera viewpoint, and you can reposition that camera in the 3D world. So, you would position the camera to get the first frame, and then you move the camera a tiny bit to the side and angle it a bit, then you get the 2nd frame. So I think the 3D depth data is just part of the model/world, and you are just changing the camera angle/position for the different frames. Or, alternating between 2 cameras.

 

[serpretetsky]

What is 2D+Depth?

Crysis 2 is, by definition, is a native stereoscopic 3D video game. What this means is that the actual game is creating the 3D experience, and it’s not an extrapolation of 3D API calls created by a third party stereoscopic 3D driver.


What makes Crysis 2 different from its 3D gaming predecessors is it is using a 2D+depth technology. Normally, a stereoscopic 3D image requires two distinct camera views which contain unique information. On paper, this could require as much as double the processing power of a traditional 2D game.

2D+depth creates a 3D experience by placing objects at different depths on a pixel by pixel basis. It works by mixing the image captured from a single camera perspective, and combining it with the information from the Z-axis to determine depth placement. The result is then passed on as a separate left and right image for the display.


2D+depth is advantageous because by only requiring a single camera view, it maintains the game’s full performance in both 2D and stereoscopic 3D modes. Unfortunately, unlike a true stereoscopic 3D game with two unique camera views, 2D+depth tends to lose a lot of bite in the process. Limitations usually include depth-only experiences, and object 3D placement and shaping is less convincing.

Crysis 2 in 3D! (Updated!)
Published on Wednesday, 20 April 2011 10:59
Written by Neil Schneider
By Neil Schneider
http://www.mtbs3d.com/index.php?option=com_content&view=article&id=12429&Itemid=76

Thank you for that omeds! I wonder what approximations they use

As far as the rendered frame vs the depth data, maybe you can think of it as you have a 3D movable camera viewpoint, and you can reposition that camera in the 3D world. So, you would position the camera to get the first frame, and then you move the camera a tiny bit to the side and angle it a bit, then you get the 2nd frame. So I think the 3D depth data is just part of the model/world, and you are just changing the camera angle/position for the different frames. Or, alternating between 2 cameras.

yes, as far as i know that's how games (and the drivers) produce the stereoscopic image.

 

[SirPauly]

Indeed --2d+depth! Didn't offer the high depth quality as traditional stereo 3d methods but the performance hit was low and very compatible with the features.