RE: 3D SPEX (and LCS in general)

[P3D <URBANIC@xxxxxxxxxxxxxx>]

OK, I guess I feel oblidged to respond here, as I'd like to make sure the
"3D Theory" pages are as clear as possible (thanks to the many that wrote).

Greg Marshall states:

>  But I'll repeat what I said before:  Sync doubling is no magical solution
> - you can describe it a few different ways but the bottom line is exactly
> what you said:  ...you keep the total bandwidth the same.  You don't want it
> the same - you want it to double (same b/w for each eye).  In other words,
> sync-doubling gives you only half the vertical resolution. Now this is fine
> and very useful for some applications, but it's no break-through technology.

I'm afraid I haven't seen any of your ealier statements regarding sync-
doubling, but I've never claimed sync-doubling is "magical", just the best
current solution, as indicated by companies that offer all the common
solutions (3DTV or Stereographics), and yet use SD in thier premium products.
It is the only technique of those that are commonly discussed here that never
suffers from flicker, and hence the only one I'd consider truly usable.
It's definitely no "break-through" technology as it has been heavily used in
serious applications for years.  For example, you may have noticed on TV that
the NASA guys all use StereoGrpahics CrystalEyes glasses for their 3D work.
Note that this is not the cheapo SimulEyes page-flipping/interlaced product
that SG tried to sell to the generic comsumer.  I would challenge anyone to
tell me that StereoGraphics and 3DTV do not own the vast majority of the NASA,
Government Lab, Nuclear, Military and University market for real stereo
applications.  Granted, these are the we-need-the-best-cost-is-no-object
customers, but I wasn't claiming that sync-doubling was the cheapest, just
the highest quality.  Greg may want to point out other techniques used for
commercial applications that he is familiar with.

As for the tradeoff in vertical resolution.  I should make two points before
anyone gets too confused.  First, the total image resolution remains the
same as for a 2D picture, say 1024x768 (the minimum resolution we use for
educational materials).  Each eye only sees half of the vertical resolution,
but the total amount of information that your eyes/brain get to see is
1024x768 truecolor pixels, giving at least the image quality you would
expect from a 1024x768 2D image (and that's not bad).  Now, many papers
claim that the image quality is actaully greater as you brain is better
processing 3D material than 2D, but that strays into subjectivity.  Just
remember that there are a total of 1024x768 pixels and you understand the
image quality.  BTW, note that you can always switch to even higher resolution
standard graphics modes if you want more resolution.  Sync-doubling allows
you to use the higest res mode your video card can put out.  Contrast this
to most of the low-res modes that the page-flipping schemes need to make
sure you monitor doesn't exceed its specs.

The second fascinating (at least to me) point I'd like to make is that
vertical resolution is less in the eye than horizontal resolution anyway.
"What is he talking about!?!", you say.  Simple, take a fine graph of
parallel bars (our optical engineer carries around such a card all the time
with differant size grids to test "line-pairs" of resolution on our stuff;
you can print one out on your printer if you want). Hold the bars so that
they are running vertically and move them away from your eyes until they
just begin to blur together, but you can still distinguish them.  Now,
rotate the bars 90 degrees so that they are running horizontal.  Can't
distinguish them anymore, right?  Pretty stunning proof that your eyes
can see better resolution in the horizontal direction.  Although there is
no concensus, most researchers believe that this is because we see
stereoscopically and hence need to distinguish finer detail in that
axis to accurately measure distance.  So, given a number of square pixels
to use for a stereo pair, it is batter to retain all horizontal resolution
than to do something like side-by-side.  It kills me to see
lossy formats like JPG used for stereo images as they lose the high-
frequency information that makes up the vertical edges.  DON'T USE LOSSY
COMPRESSION FOR 3D.  Anyway, I'm always amazed at how many "stereoscopic"
companies aren't even aware of basic phenomonae such as this and will go
ahead and invest millions in a product.

Greg:
>> (2) Stereo images on CRTs exhibit ghosting due to the
>> long exponential decay of the phosphors.

Me:
>OR, and this is a big OR, reprocessing the image to mute the problem
>chromal boundries.

Greg:
>Agreed.  I've done some testing of this technique with less than
>spectacular results, but I haven't had that much time to work on it.

It took a _lot_ of time for us to develop these algorithms, and we generally
think that they are our most valuable asset.

> Of course, it detracts from the overall dynamic range of the image and
> requires considerable processing for each image.  How well does your
> technique work with varying phosphors?  I've encountered quite a range in
> differential decay rates on various monitors.

It only detracts from the overall dynamic range if you do something crude,
like a gross gamma correction (or dark lenses on the glasses!); if you
correct only across the boundries, this does not need to occur.  We spent
a lot if time and reseach on monitor phosphors, and most companies (and
there aren't as many tube companies as you think) seem to use very similar
phosphors.  This is good and bad.  They all have horendous asymetries in
response between the phosphors, but they are similar in this respect.

Greg:
>> CRTs will eventually be replaced by various flat panel displays
>> which (guess what?) are even more problematic for stereo!

Me:
>If these were LCD's you would be right, but the future at this point seems
>to be all Digital Micro Mirror

Greg:
>Good point, I'd forgotten about DMM.  This is important because this
>technology will probably replace commercial film projectors.  However,
>considerable advancement and cost reduction is needed before it can
>be used for "personal" displays.

It has already, in the year since commercial release, taken over much of
the commercial projection market, and become the premium choice in the
home theater market.  Not bad.  And because it is just a silicon based
chip, we should expect to see continuing price reductions like we see for
all other such devices, whereas large plasma and LCD displays have been
under intense development for years and continue to face the same fundemental
large-scale problems.  Never bet against silicon.

> These devices already use an enormous
> amount of bandwidth to modulate intensity via time division - doubling
> that to provide stereo would undoubtedly cause some TI engineers to
> pull their hair out in frustration!

All the bandwidth is on the chip, where more-than-enormous amounts are
available (even your PC has a multi-Gigbit bus).  Also, TI seems to be
quite pleased with the stereo that they are demoing (sync-doubling, I
might add).  So are some of our customers.

And, as I feared, Greg's explanation of sync-doubling has already confused
some attentive listeners:

Larry Berlin writes:
>******  Not only does synch doubling NOT provide higher resolution images,
>it uses images with half the vertical height. That's decidedly an unwanted
>side effect... Half height images are not natural or easy to look at...

Larry, rest assured that no sync-doubling images are half height.  Any
monitor automatically stretches out any frame to fill the full screen.
Every one of our images, for example, looks at least like a nice, bright,
crisp, flicker-free 1024x768 image, but in 3D.

Larry does mention a very good point:
> Besides, any good computer needs hardware acceleration for
> handling todays' graphic intensive environment.

That is why sync-doubling works so well.  By using normal graphics modes,
and the original drivers for the video cards, sync-doubling allows you to
use the latest greatest (or old, but faithful):

                              video/accelerator cards
                              DirectX game technology
                              Windows 95
                              Windows NT
                              Digital Video Disk

We had a Digital Video Disk product weeks after they hit the market.  I
don't think any of the other techniques even work with Windows NT, let
alone DVD, and there are pretty good technical reasons to believe they can't.

> Increasing it's power to handle stereo requirements provides a much truer
> improvement to stereo viewing that is extremely cost effective. And the cost
> is very comparable

I think Larry is trying to say that the video cards should provide stereo
on-board.  If so, I couldn't agree more, and it wouldn't only be comparable
in cost, it would be much cheaper.  Unfortunately, it not only isn't going
to happen, the world is going the other way.  Video cards used to support
a feature called Vertical Blanking Interupt which made real page-flipping
possible.  The producers have dropped it, making it impossible to do outside
of DOS, where you can fake it.  There have always been a few vendors producing
cards that can support stereo directly (remember Kassan?), but they have
always been such a small minority that the installed base of stero-capable
cards has been shrinking as the older VBI cards obsolete.  We really only
wanted to be an educational software company ourselves, but you can't make a
product that only 5% of the installed base can use, so we had to develop
our own hardware (or sell $600 CrystalEyes, something most schools can't
afford).

I find it a bit ironic that the very small stereo enthusiast market waits
for the huge 2D video hardware world to accomodate them, which has never shown
the least sign of happening.  Meanwhile, the one thing that they do have
control over, a software spec for stereo images, they adopt probably the
least applicable 2D spec there is (JPEG, a lossy compression scheme with no
registration information).  If the dozen dead and dying stereoscopic hardware
companies that litter the past 5 years don't convince you (you'll see 2 more
major failures this coming year), I don't know what will, but hopes like:

>******  While this might work in some cases, it wouldn't work in all cases
>and represents FAR more work than is necessary. Here again it's a better
>approach to improve both monitor screens and the operation of LCS glasses.

for the hundred billion dollar monitor indistry, and

>>*****  They should be starting with the spec for stereo image work as the
>basis of development. Then not only would it be robust enough for stereo, it
>would have no problem with regular applications either. Doubling a process
>shouldn't be too difficult

for the enourmously powerful video spec commitees are way too optimistic.
These people can't agree on important economic issues like copyright 
protection.  Do you really think they care in the least about our little
corner of the world?  Continue to wait if you must, but I suggest going
with something that works well today.  I find it a bit ironic that there
are more Biology teachers using high-quality electronic stereo and taking it
for granted than there are members in the NSA.

Wow, this got long.  On the other hand, it beats typing it 20 differant ways
in response to e-mail (but please feel free to e-mail).

John


------------------------------