Digital Foundry Article Technical Discussion Archive [2010]

[Arwin]

But that would be 640 + depth, not 1280 + depth. The depth information can only be 'decoded' when those lines refer to the same information at a slightly different angle, not if each eye gets a different line altogether.

 

[Shifty Geezer]

I still don't think there's a good way to do it. After all imagine how games that have attempted to display explosions using 3D polygons. Not so good. Except now you'd have to take a 2D texture and try to make a very irregular polygon out of it. So suddenly your triangles per scene will explode, and you'll have to reduce resolution or scene complexity even further.

Nope, that's not what I 'm suggesting. There's zero extra geometry for rendering explosions.

You understand the idea fo 3D from 2D right, as used in Crysis 2 and Sony's tech? You render one single viewpoint, and then derive 2 from it by shifting pixels around based on depth. That's a single image, you could take a photograph of someone's head, and use depth info to extrude it in 3D space. That's all it needs, so if you render a flat polygon but include in the texture some depth info, you'll have what's needed to create 3D from 2D by shifting pixels around, only with the pixel displacements not just being on the flat plain of the polygon resulting in a two dimensional object in the 3D space, but instead they'll be displaced as if the centre of the polygon is at a different distance. In the case of our explosions, each particle can have a bulge in its centre like a displacement map, and the 3D from 2D process will make it look 3D within the boundaries of the object. Thanks to explosion clouds being pretty much spherical and somewhat hazy, it should be quite convincing.

The overhead would be storage for a depth map per particle (insignificant) and the 3Dification process which is that 2% overhead that CryTec have talked about. For just rendering explosions, that's a significant overhead, but if other aspects of the engine are reigned in a bit, it seems doable, and something probably necessary as 2D sprites are going to look awful and break the illusion.

 

[Silent_Buddha]

Nope, that's not what I 'm suggesting. There's zero extra geometry for rendering explosions.

You understand the idea fo 3D from 2D right, as used in Crysis 2 and Sony's tech? You render one single viewpoint, and then derive 2 from it by shifting pixels around based on depth. That's a single image, you could take a photograph of someone's head, and use depth info to extrude it in 3D space. That's all it needs, so if you render a flat polygon but include in the texture some depth info, you'll have what's needed to create 3D from 2D by shifting pixels around, only with the pixel displacements not just being on the flat plain of the polygon resulting in a two dimensional object in the 3D space, but instead they'll be displaced as if the centre of the polygon is at a different distance. In the case of our explosions, each particle can have a bulge in its centre like a displacement map, and the 3D from 2D process will make it look 3D within the boundaries of the object. Thanks to explosion clouds being pretty much spherical and somewhat hazy, it should be quite convincing.

The overhead would be storage for a depth map per particle (insignificant) and the 3Dification process which is that 2% overhead that CryTec have talked about. For just rendering explosions, that's a significant overhead, but if other aspects of the engine are reigned in a bit, it seems doable, and something probably necessary as 2D sprites are going to look awful and break the illusion.

I understand what you're saying but I remain unconvinced. I guess we'll have to wait and see how they end up dealing with the issue and how convincing their work arounds turn out.

Regards,
SB

 

[Silent_Buddha]

For 720P/2 3D (KZ3 style)

left eye: 640 vertical lines (set a)

right eye: 640 vertical lines (set b)

brain: 1280 unique lines (set a + set b) + depth information



But I think I understand what you mean.

I thought he meant effective vertical interlace due to line doubling.

For the 640 veritcal lines they aren't interlaced to provide 1280 combined vertical lines. They are the exact same vertical lines except shifted to the right or left depending on which eye that frame is meant for.

An interlaced scene where odd lines are shifted to the right and even lines are shifted to the left would be bizarre in the extreme. And when using glasses to then provide only even lines to one eye and only odd lines to the other eye would be even more bizarre.

Regards,
SB

 

[Laa-Yosh]

Besides, vertical rows of pixels are called columns, usually. Only a horizontal row is called a line... at least that's the most commonly used naming convention that I've heard of.

 

[ihamoitc2005]

For the 640 veritcal lines they aren't interlaced to provide 1280 combined vertical lines. They are the exact same vertical lines except shifted to the right or left depending on which eye that frame is meant for.

An interlaced scene where odd lines are shifted to the right and even lines are shifted to the left would be bizarre in the extreme. And when using glasses to then provide only even lines to one eye and only odd lines to the other eye would be even more bizarre.

Regards,
SB

I disagree, my friend, the pixels are completely different, no? Yes, on paper it is 640 lines but for the brain it sees much more because each eye gets different information.

Left eye: 640 columns (set a)

Right eye: 640 columns (set b)

Brain: 1280 non-repeating columns (set a + set b) + depth information

 

[SonComet]

I disagree, my friend, the pixels are completely different, no? Yes, on paper it is 640 lines but for the brain it sees much more because each eye gets different information.

Left eye: 640 columns (set a)

Right eye: 640 columns (set b)

Brain: 1280 non-repeating columns (set a + set b) + depth information

That is incorrect. I had friends at E3 that confirmed that 3D KZ3 looked very very low resolution during the press conference and on the show floor. They preferred the 2D mode and these are people that absolutely love 3D movies.

 

[dumbo11]

Looks like Sony is also working on getting the "fake" 3D up to acceptable levels ala Crysis 2. Implication being that since there's less of a performance hit, they might be pushing that in the future rather than "real" stereoscopic 3D.

AFAIK the "fake" 3d has an important limitation to go with the good points:

"Crysis 2 uses a concave 3D effect so that objects do not pop out of the screen, which gives the game “more depth” according to Yerli."

I assume that the closer an object appears to the viewer, the more it appears to be a cardboard cut-out [as it lacks left/right information for itself, and increases the size of the 'black holes' exposed it?].

In KZ3 bullets will appear to 'fly past your face' - in Crysis 2 the bullets will appear to 'hurtle towards the screen'.

It will be interesting to see how that works out - I guess "real 3d" might get tiring after a bit? maybe there's a method to switch from one to the other for 'cinematic events'?

 

[patsu]

They should be familiar with multiple techniques to suit different performance profiles. If the developers can switch between 3D techniques on-the-fly, they may be able to optimize the final experience and their development resources better.

If people enjoy 3D gaming on 3DS and/or PS3, then it will stay and proliferate to other devices. Otherwise, it'll be forgotten. If they execute well, the adoption will take a few years, or even decade.

 

[Silent_Buddha]

I disagree, my friend, the pixels are completely different, no? Yes, on paper it is 640 lines but for the brain it sees much more because each eye gets different information.

Left eye: 640 columns (set a)

Right eye: 640 columns (set b)

Brain: 1280 non-repeating columns (set a + set b) + depth information

No that would be the same as saying the 2D version with 1280 columns would equal 2560 columns for 2 successive frames.

It doesn't work like that. Your left eye will see 640 columns. Your right eye then sees its 640 columns superimposed directly over where the 640 columns were for the right eye.

Your brain will only ever see 640 columns. Your TV will only ever display 640 columns (well it'll be upscaled to 720p by the TV or console, but you get the idea). So after upscaling each pixel in game will take up 2 pixels on screen (assuming 720 horizontal lines). The image itself still remains 640 columns, albeit stretched out to 1280.

Regards,
SB

 

[ihamoitc2005]

No that would be the same as saying the 2D version with 1280 columns would equal 2560 columns for 2 successive frames.

It doesn't work like that. Your left eye will see 640 columns. Your right eye then sees its 640 columns superimposed directly over where the 640 columns were for the right eye.

Your brain will only ever see 640 columns. Your TV will only ever display 640 columns (well it'll be upscaled to 720p by the TV or console, but you get the idea). So after upscaling each pixel in game will take up 2 pixels on screen (assuming 720 horizontal lines). The image itself still remains 640 columns, albeit stretched out to 1280.

Regards,
SB

What you miss, my friend, is that when you "super-impose" these columns (let's pick random column 355), the brain processes the fact that the pixel in left eye column 355 and the pixel in right eye column 355 are different and may even describe different sections of the scene (due to distance between eyes/viewpoints), and that the details in adjacent colums (even more important because view is off-set) are also different (context), and extracts both detail and depth to recreate mental image and understanding. This is because although they are both column 355, they are describing different parts of the scene as well as different objects withing the scene.

With stereo vision, any matching column may describe slightly different sides of the same object but also describing different parts of different objects at the same time. Column for one eye may even see something that is not visible on the same column of the other eye. But the brain will see it.

The reality is that while a 2D image, both eyes see the same point of detail for every object in every column, in 3D, no column provides the same detail or the same object.

For 2D processing, both eyes see the same thing. Column 355 from both eyes are exactly the same. No additional information is provided to the brain.

 

[dumbo11]

No that would be the same as saying the 2D version with 1280 columns would equal 2560 columns for 2 successive frames.

It doesn't work like that. Your left eye will see 640 columns. Your right eye then sees its 640 columns superimposed directly over where the 640 columns were for the right eye.

Your brain will only ever see 640 columns. Your TV will only ever display 640 columns (well it'll be upscaled to 720p by the TV or console, but you get the idea). So after upscaling each pixel in game will take up 2 pixels on screen (assuming 720 horizontal lines). The image itself still remains 640 columns, albeit stretched out to 1280.

Regards,
SB

series != parallel

In 2d, the viewer is shown 1280 columns - it is important for the viewer to perceive that image as 'now', and the subsequent 1280 columns are now+0.1 seconds or whatever.

The purpose of 3d is to make the viewer believe that image L and image R are the same scene from different viewpoints. If 3d is to work, then the viewers eyes would need to perceive the equivalent of a '2560' pixel wide stereo image to your brain.

I'd guess that your 'visual cortex' (or whatever it's called) decides that much of the L image is duplicate data from the R eye, and so what you end up perceiving is somewhere between 1280 and 2560 with depth added.

 

[Shifty Geezer]

Though there's an argument to be made that 3D is less...'dependent' on resolution, I don't think these numbers are the way to go about it. Presented two pieces of data on the same information, left and right views of a scene, the human mind will interpolate all sorts of detail - most of what you 'see' in a given moment is actually all made up by the brain. HD resolutions are about fidelty and giving a scene some clarity. It also reduces artefacts like aliasing. 3D will, in my estimation, increase information density and so decrease perception of jaggies, but not affect fidelity. The end result is 3D at a given resolution probably looks less aliased than that same resolution in 2D, but won't look as sharp.

The approach looking at numbers and samples isn't an effective way to explain human perception. I imagine there are some good research papers that explore this, but I don't know any!

 

[ihamoitc2005]

Though there's an argument to be made that 3D is less...'dependent' on resolution, I don't think these numbers are the way to go about it. Presented two pieces of data on the same information, left and right views of a scene, the human mind will interpolate all sorts of detail - most of what you 'see' in a given moment is actually all made up by the brain. HD resolutions are about fidelty and giving a scene some clarity. It also reduces artefacts like aliasing. 3D will, in my estimation, increase information density and so decrease perception of jaggies, but not affect fidelity. The end result is 3D at a given resolution probably looks less aliased than that same resolution in 2D, but won't look as sharp.

The approach looking at numbers and samples isn't an effective way to explain human perception. I imagine there are some good research papers that explore this, but I don't know any!

On this I agree, my friend. You can't simply say one column is super-imposed over other and so it's all the same lines.

The brain recreates a new image (like dreaming) based on all the different data it receives. It is the ultimate GPU! Only problem ... no HDMI output. It will be great to connect a DVR at night.

 

[Arwin]

On this I agree, my friend. You can't simply say one column is super-imposed over other and so it's all the same lines.

The brain recreates a new image (like dreaming) based on all the different data it receives. It is the ultimate GPU! Only problem ... no HDMI output. It will be great to connect a DVR at night.

I still say you're wrong. Let's for instance imagine that you're looking at a 2D picture on a 2D screen. Now we're going to show that 2D picture on the 3D screen, but the picture stays in 2D. All we're doing is placing this 2D picture slightly in front of the physical screen's edges, at exactly the same level as the screen's edges, or slightly into the screen. The picture is still 2D. It's the same resolution it was before. You see it in the same resolution as before. Your brain interprets it the same way. Except, now it is perceived to be at a different distance from you relative to the edge of the screen (or not).

Now we should of course not undervalue the third dimension that has just been added. But you're mistaken if you believe any x or y information is added - only z information is added. The source screen still only encodes half the x or half the y information, and this information doesn't magically return. Pixel aliasing just has an additional dimension to be visible in, and it is in fact suggested from various sources that this helps your brain identify the individual pixels / aliasing steps more accurately (you see them now in 3 dimensions) rather than the other way around, which is why Sony's researchers say that good AA is more important than resolution. I presume this is because your brain gets even better at edge detection in 3D space, which makes sense if you think about it.

 

[ihamoitc2005]

I still say you're wrong. Let's for instance imagine that you're looking at a 2D picture on a 2D screen. Now we're going to show that 2D picture on the 3D screen, but the picture stays in 2D. All we're doing is placing this 2D picture slightly in front of the physical screen's edges, at exactly the same level as the screen's edges, or slightly into the screen. The picture is still 2D. It's the same resolution it was before. You see it in the same resolution as before. Your brain interprets it the same way. Except, now it is perceived to be at a different distance from you relative to the edge of the screen (or not).

Now we should of course not undervalue the third dimension that has just been added. But you're mistaken if you believe any x or y information is added - only z information is added. The source screen still only encodes half the x or half the y information, and this information doesn't magically return. Pixel aliasing just has an additional dimension to be visible in, and it is in fact suggested from various sources that this helps your brain identify the individual pixels / aliasing steps more accurately (you see them now in 3 dimensions) rather than the other way around, which is why Sony's researchers say that good AA is more important than resolution. I presume this is because your brain gets even better at edge detection in 3D space, which makes sense if you think about it.

Because each eye sees completely different images (no shared pixels) the X + Y information is also added. That is why all 1280 columns are unique.

The depth is because of stereoscopic parallax between eyes.

This is because corresponding columns describe different objects or different sides (adjacent details) of the same object.

If you draw on a sheet of paper from overhead view and draw sight lines to different objects and different depths (maybe even some blocking) you will see that it is because x and y are different that you brain interprets z.

 

[Laa-Yosh]

The difference between pixels is like 5% on average...

 

[Sigfried1977]

A 640p resolution does not look like a 1280 resolution with glasses. It looks like a 3d 640p image and that's it. Tried this a couple of times with my PC.
What's definitely true is that resolution isn't that big of a factor in 3d anymore. Neither is stuff like great black levels.

 

[ihamoitc2005]

5%?

The difference between pixels is like 5% on average...

What is the difference between pixels in 2D?

 

[ihamoitc2005]

A 640p resolution does not look like a 1280 resolution with glasses. It looks like a 3d 640p image and that's it. Tried this a couple of times with my PC.
What's definitely true is that resolution isn't that big of a factor in 3d anymore. Neither is stuff like great black levels.

The disparity is because when you look at the image to analyze you end up favoring one eye (test and you will see) so you end up focusing on a left or right (depends on individual) 640 column image but when you play it combines and you see everything and more.