Digital Foundry Article Technical Discussion Archive [2010]

[TrungGap]

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.

I don't quite understand what you mean by "no shared pixels". I thought the idea behind 3D was to fool your brain by showing a different pixel at the same location, so you perceive depth. If your brain interpreted the two images as spatial resolution, then no depth information could be derived.

 

[Laa-Yosh]

Also, if there was no quality difference between 640x720 at 3D and 1280x720 at 2D then no-one would ever look for alternate solutions to provide two 1280x720 images even in stereo. But there's some serious Sony research going on.

 

[ihamoitc2005]

I don't quite understand what you mean by "no shared pixels". I thought the idea behind 3D was to fool your brain by showing a different pixel at the same location, so you perceive depth. If your brain interpreted the two images as spatial resolution, then no depth information could be derived.

It recreates internal model of object based on different images of the same objects.

Similar to how if you walk around a car you now know how it looks like from all sides (including some brain filling in blanks such as undercarriage) and can now draw the object from any angle.

3D artists do this all the time.

Of course, in real life, when we view real objects, there is also continuous jitter (eye and body movement) and also unique vertical displacement of eyes (most people, one eye is higher than the other to varying degreees) and this is not possible to provide with 3D TV such that it matches with each viewers brain orientation (due to specific placement of eyes and movements).

 

[ihamoitc2005]

More resolution is always better.

Also, if there was no quality difference between 640x720 at 3D and 1280x720 at 2D then no-one would ever look for alternate solutions to provide two 1280x720 images even in stereo. But there's some serious Sony research going on.

And 640x720 is very very bad for screen shots.

In 3D stereo it is enough for action games where 1280x720P is enough.

For example, your brain can catch one pixel movement in one eye at 30hz. This is shown in many tests. One test claims an Airforce pilot caught one pixel at 1/500th second.

So can you see distant changes (new enemy or popping out) just as well as long as you are close to the TV and you see with sufficient concentration of your retina rods.

Big problem is I hate 3D glasses.

 

[Laa-Yosh]

This is just your own personal idea based on some numerology and several guys have already expressed their opinion that the theory does not work in practice. Brains and pixels and whatever.

 

[Shifty Geezer]

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.

Let's be clear on what resolution is, and hopefully you'll see this is wrong. Our displays present an image as an organised array of light values. The scene is mathematicallly modelled in 3D and then transformed, projected and resolved, sampling discrete points of the scene to get the light value. At a resolution of 300x200, there will be 60,000 samples. If you render the same scene now at 600x200, the new samples lie in between the others. You are harvesting information from the same scene, the same projection. You could also render 300x200 pixels, then move the camera a half pixel to the right and render 300x200 pixels, that information corresponding to the even columns of the 600x200 image, and then interweave the two sample sets to generate an identical 600x200 image, again both odd and even columns coming from the same scene, same projection.

However, if you sample half the pixels from one camera, and the other pixels from a different camera looking at a different scene, you have broken the ordered grid deconstruction of the image. As an extreme case to prove the point, consider rendering 300x200 of a camera looking down a street in GTA4, and then 300x200 pixels inside a bar in GTA4, and then interleave the columns of those two to make one 600x200 image. It's not going to be an actual 600x200 image! Now hypothetically move the camera nearer, so the one down the street moves towards the bar entrance, while the one in the bar moves nearer the door. Combine two 300x200 images from these cameras and you still don't get a single, uniform sampling of the same scene. Now move the cameras to within one virtual metre...still not a single image. Now place them next to each other, 3" apart, looking at slightly different directions. The result remains two discrete scenes sampled from two different points creating two separate sample sets, that cannot be combined to create a single higher resolution image.

Thus 3D cannot be considered the same resolution as 2x the horizontal res of each eye image. 3D will be perceived differently, meaning comparisons of resolution between 2D and 3D won't be the same, and 1280x720 may look sharper or less sharp than 2x 1280x720 in 3D (and it would appear from what Arwin posted that my original guess is back-to-front, and 3D will increase sensitivity to resolution, which actually makes more sense. The whole is worth more than the sum of the parts and all that), but in actual metrics, 3D resolutions cannot be considered the same as the doubled-up 2D resolution of the same aggregate dimensions.

 

[ihamoitc2005]

Everyone has an opinion, my friend.

This is just your own personal idea based on some numerology and several guys have already expressed their opinion that the theory does not work in practice. Brains and pixels and whatever.

There is no fact to be found on this subject because neuroscience is not such a field. Life is like that.

There are many who cannot live without the illusion of mathematical fact so for them I recommend the following:

http://www.sanatansociety.org/vedic_astrology_and_numerology/calculate_your_numbers.htm
http://www.pythagoreannumerology.com/
http://www.suite101.com/lesson.cfm/19170/2776

 

[ihamoitc2005]

Yes

The whole is worth more than the sum of the parts and all that), but in actual metrics, 3D resolutions cannot be considered the same as the doubled-up 2D resolution of the same aggregate dimensions.

This is my point.

It is not the same.

The brain perceives 3D data very differently. A 1280x720 3D image (2x640x720) provides the brain with far more information than a 1280x720 2D image.

It's not super-imposed columns. Brain doesn't work like that.

 

[mrcorbo]

This is my point.

It is not the same.

The brain perceives 3D data very differently. A 1280x720 3D image (2x640x720) provides the brain with far more information than a 1280x720 2D image.

It's not super-imposed columns. Brain doesn't work like that.

You're sacrificing horizontal detail, though, in the final image your brain resolves. The human visual system also tosses out a lot of the information it receives from each eye in assembling the final image you perceive. This is pretty clearly demonstrated by the fact that you can close your non-dominant eye and not automatically see 50% less detail in what you are looking at.

 

[ihamoitc2005]

1) When you close one eye, you loose a large section of your FOV/horizontal resolution. That's a big detail loss.

2) The real world isn't made of pixels. There's plenty of detail per sq inch of view no matter if you block one eye or peer through a hole in the wall.

3) If the brain was throwing out "a lot of that" data it could not get the depth/shape of objects. It is because brain sees all adjacent pixels in stereoscopic view that it is able to map those colors to shapes and also know depth.

4) Even surface texture/quality is more apparent for stereo vision. This is because for a uniform colored surface that is in view of both eyes, each eye sees a different color for due to slightly different angle of light reflection and the brain can understand this.

 

[assen]

We've always been at war with Eastasia?

 

[Sigfried1977]

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.

Does it matter? In practice it looks clearly a lot worse, and you don't have to pause and start analyzing the image for it either. The difference shows just fine in motion.

 

[ShadowRunner]

Would it not look better render two full 720p buffers but have the first image only render odd pixel columns, and the second image even? the blank columns could be filled using the colour from the adjacent pixel column so you would not see black lines through the image. Refreshing at 120hz you should see it as a full 720p image no?

 

[upnorthsox]

Would it not look better render two full 720p buffers but have the first image only render odd pixel columns, and the second image even? the blank columns could be filled using the colour from the adjacent pixel column so you would not see black lines through the image. Refreshing at 120hz you should see it as a full 720p image no?

Yes there's no reason you can't interlace, in fact that's how some 3d tv will be.

 

[ihamoitc2005]

This is what they do

Would it not look better render two full 720p buffers but have the first image only render odd pixel columns, and the second image even? the blank columns could be filled using the colour from the adjacent pixel column so you would not see black lines through the image. Refreshing at 120hz you should see it as a full 720p image no?

What you are saying is render every other column and copy pixels to fill blank columns. This is horizontal scaling and this is what they do.

They actually render 1280 unique columns with 640 for each eye (similar effect to distant column technique you mention) but because it is from two different points of view (so see different areas of same objects) you get much more understanding of surface material, shapes, distance, lighting, shadows, etc. are all much more clear and realistic. Only texture is a little bit less clear than 1280 column 2D.

 

[Silent_Buddha]

Would it not look better render two full 720p buffers but have the first image only render odd pixel columns, and the second image even? the blank columns could be filled using the colour from the adjacent pixel column so you would not see black lines through the image. Refreshing at 120hz you should see it as a full 720p image no?

That would be impossible. Well not impossible but the result would be very unpleasant.

The reason 2D video streams can be interlace is that there is generally when things change, they change predictably as the eye expects. Frame A -> Frame B -> Frame C -> Etc. where Frame B is a natural progression of Frame A and Frame C is a natural progression of Frame B.

That makes it relatively easy for TVs to de-interlace the scene into what your eyes expect to see.

With 3D and alternating frames, Frame A <> Frame B. In fact, what you get is Frame A/Frame B -> Frame C/Frame D -> Etc.

Note that I didn't use Frame A + Frame B and Frame A doesn't not lead to Frame B.

In this case, Frame A (one eye) and Frame B (other eye) are 2 very distinctly different frames. Each has to be displayed seperately from the other and there can be no combining of the two. Each is meant for a specific view point.

If you tried to do odd lines for one eye and even lines for the other eye, you would end up with a striped scene rather than a de-interlaced scene.

Whatever fantasies some may hold. A stereoscopic scene where left eye has X lines of resolution and right eye has X lines/columns of resolution, when viewed in motion you will not see a scene with X+X lines/columns of resolution.

Both science AND actual hands on experience will show that. Depending on how a scene is constructed, you may or may not have some non overlap giving you a 2D extension/border beyond the X lines/columns of resolution but you'll never have 3D resolution greater than X.

Anyone expecting KZ3 in 3D for example to have resolution greater than the dimensions of each individual frame is going to be hugely disappointed.

Regards,
SB

 

[ihamoitc2005]

You think too much 2D

That would be impossible. Well not impossible but the result would be very unpleasant.

The reason 2D video streams can be interlace is that there is generally when things change, they change predictably as the eye expects. Frame A -> Frame B -> Frame C -> Etc. where Frame B is a natural progression of Frame A and Frame C is a natural progression of Frame B.

That makes it relatively easy for TVs to de-interlace the scene into what your eyes expect to see.

With 3D and alternating frames, Frame A <> Frame B. In fact, what you get is Frame A/Frame B -> Frame C/Frame D -> Etc.

Note that I didn't use Frame A + Frame B and Frame A doesn't not lead to Frame B.

In this case, Frame A (one eye) and Frame B (other eye) are 2 very distinctly different frames. Each has to be displayed seperately from the other and there can be no combining of the two. Each is meant for a specific view point.

If you tried to do odd lines for one eye and even lines for the other eye, you would end up with a striped scene rather than a de-interlaced scene.

Whatever fantasies some may hold. A stereoscopic scene where left eye has X lines of resolution and right eye has X lines/columns of resolution, when viewed in motion you will not see a scene with X+X lines/columns of resolution.

Both science AND actual hands on experience will show that. Depending on how a scene is constructed, you may or may not have some non overlap giving you a 2D extension/border beyond the X lines/columns of resolution but you'll never have 3D resolution greater than X.

Anyone expecting KZ3 in 3D for example to have resolution greater than the dimensions of each individual frame is going to be hugely disappointed.

Regards,
SB

2D is all about what hits the eyes. This is ok because both eyes get exactly same information.

3D is all about what brain perceives based on the fact that each eye gets completely different information.

Whole new way to think.

 

[liolio]

Put a pile of sugar dices in front of you (a pyramid), look with one eye the other and both, do you think that when looking with both eyes there is more sugar dices? Or they are twice as tinier?
I guess not, no replace sugar dices by pixels and you should understand that what you say make imho no sense.

 

[Neb]

Doesn't mather, point is it will basically be as sharp as 640x720 frame would be. It will not look as sharp or clean as a 1280x720. There is no way to spin it around.

 

[Arwin]

2D is all about what hits the eyes. This is ok because both eyes get exactly same information.

3D is all about what brain perceives based on the fact that each eye gets completely different information.

Whole new way to think.

I still think you're exaggerating the differences. 3D is all about the fact that each eye gets the same information, but with just a slight enough difference so that it can perceive 3D. Had the information been completely different, it wouldn't have been able to perceive depth.

Of course there is a lot more going on in actual 3D spatial awareness, but just seeing distance accurately as simulated by the 3DTV technology that we're discussing now isn't that complicated, and you're turning it into something more than it actually is.