Newsgroups: sci.image.processing
From: paul@pcserv.demon.co.uk (Paul Carpenter)
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!news.mathworks.com!hookup!news.sprintlink.net!peernews.demon.co.uk!pcserv.demon.co.uk!paul
Subject: Re: Matrox vs Data Translation - Opinions?
Distribution: world
References: <huizen.792336001@appliedmicro.ns.ca> <3hhrgh$5ec@taco.cc.ncsu.edu> <792591509snz@pcserv.demon.co.uk> <3hlpsl$70h@taco.cc.ncsu.edu>
Organization: PC Services
Reply-To: paul@pcserv.demon.co.uk
Followup-To: sci.image.processing
X-Newsreader: Demon Internet Simple News v1.29
Lines: 208
X-Posting-Host: pcserv.demon.co.uk
Date: Mon, 13 Feb 1995 09:31:51 +0000
Message-ID: <792667911snz@pcserv.demon.co.uk>
Sender: usenet@demon.co.uk

sggoodri@eos.ncsu.edu "Steven Georg Goodridge" writes:
-paul@pcserv.demon.co.uk (Paul Carpenter) writes:
-> sggoodri@eos.ncsu.edu "Steven Georg Goodridge" writes:
-> -Our lab had bad experiences with both vendors. Data Translation's boards
-> -were over-priced, and when we ordered two of them...

I did originally ask for this issue by email, so my answers will be 
exceptionally brief on the aspect of a particular order with one vendor!!

-We bought two DT2871 color frame grabbers from Data Translation, with DOS 
-image processing libraries, at $5000 each.

A SEVEN year old design and libraries that are *mainly* meant for getting
over the problems of hardware interfacing to driver and operating system.

-We waited two months for the hardware and three months for the software.

That surprises me on the software, and the hardware is down to old parts

-The tutorial software bundled with the hardware did not work properly.
-When we called DT about this, the technician said he was not surprised.
-They had little faith in their own product.

To my knowledge last updated FOUR years ago..

->... For some vendors it is difficult to decide which way to go, you have your
-> choice others may want AIX/AS400 support, or NT or other 32 bit systems.

-Once we gave up on DOS in favor of a  32-bit OS for the PC, we chose the 
-most popular one: OS/2. There are 10 times more copies of OS/2 in use than
-Windows NT.  Sure, there are reasons for using the other platforms. But
-our requirements are not so esoteric as one might think.

Maybe at the time it was most popular but does not mean it will be for next 
3 years, that is the problem facing vendors. Please remember that the vast 
majority of PC peripherals of all types are still sold and supported on 16 bit 
ISA bus and have to primarily work 16 bit DOS and Windows.

-> -It is ironic that both of these vendors built special hardware in ...
->.. the last year to 18 months there has been a plethora of busses, operating 
-> environments and vapourware systems.

-OS/2 is not vapourware.

That was *not* stated, however a lot of Chicago, Cairo, Windows 95/96/97.. have
appeared to have been very much this way.

-There is a huge installed base. It has existed for
-years now. Its API is stable. But the masochistic vendors worship Microsoft....

Because of the platforms that will change to 32 bit the much larger existing
base that is likely to change to 32 bit soon is already with Microsoft, so will
very quickly become the largest installed base. Whether you like Microsoft or
not, *personally* I have not been too pleased with some utopian offerings.
Unfortunately that at the moment is what MOST customers for cards ask for.

->.. So it has been difficult to choose
-> which bus to use to get better hardware throughput (PCI/VL local bus
-> /EISA/PCMCIA I or II or III...)..... etc..), all of which create a
-> lot more software writing, testing and COST.

-The OS determines how to compile the application software. It is possible to 
-write the basic software to be OS-independent (even the GUI if you are smart)
-and re-compile it for different OS's.

Would be nice but *some* of the capabilities that users require for some aspects
are *not* the same across MS-Windows, X-Windows, SunOS, Mac-Windows, Windows 
95/96....

-The BUS and OS combination determines how to write the device driver. The 
-device driver should be a separate abstraction from the system software; 
-its internals should not complex the application software. When a new bus
- technology becomes available, just change the device driver to match the 
-new hardware.
-Don't let it interfere with application development. They are separate 
-concerns.

To remind you of your original point....

-> -It is ironic that both of these vendors built special hardware in order to
-> -overcome the limits of 16 bit DOS - and this special hardware increases the
-> -cost of their products. Segmented memory limits the speed at which 
-> -image pixels in RAM may be addressed (and makes coding more complex) so
-> -they bypass this DOS limitation by doing everything on the card. But this
-> -solution is not very flexible.

Half of the 16 bit DOS limits comes from the fact that it still has to operate
on a 16 bit bus and its architecture, related to the CPU and its modes of 
operations. True 32 bit Operating systems are not much use until you can fully
remove the kludges that fix compatability with 16 bit bus and DOS based 
peripherals and memory.

These should not interfere with the application but the SPI level can be vastly
different between OS's and sometimes dependent on type of display capability
can effect interactions with application.

-What I think is insane is that these companies expect you to pay thousands of
-dollars, more than the entire cost of a PC, to buy equipment specially
-designed to overcome the limitations of DOS. Doesn't it make more sense to
-upgrade to a modern OS? Everything becomes so much easier. I just don't
-understand why these companies have long ignored the reality of OS/2.
-It is a triumph of marketing over good engineering.

Basically you are unhappy with what you got, feeling that you have been conned,
all of which is justifiable, and I think the delays are inexcusable. But you 
are comparing horse and cart with Formula 1 racings cars and expecting both to 
have the same performance. You can still buy a horse and cart as well as Formula 
1 racing cars. As a card that only works as 16 bit ISA surely it is more
liklely that is was designed to work in 16bit DOS and/or Windows?

-> Remember also that designing and selling frame grabbers is several orders of 
-> magnitude *smaller* than selling PCs or modems. So the market will be even
-> more marked in the 'get what you paid for' differences.

-Thanks to the multimedia craze, images are now a common data type. Support
-for multimedia, and video capture, is featured in modern operating systems
-like OS/2. Video capture boards can be bought in your corner PC store.

All of these ONLY support standard video formats of just NTSC related formats.

What about those working with 12/16 or greater bit depth in monochrome, 
resolutions of 1k x 1k and greater, slow and high speed clocked analog/didgital
systems. If the EVER get supported by the multimedia viewable boards it will 
take 5 years plus. Often the ability to work with multiple sources and being
programmeable is the extra cost involved with NEW designs of image processing
cards.

Remember to not compare apples and bananas.

-The PC microprocessor is getting faster than many dedicated image processing 
-boards.

NOT FOR ALL APPLICATIONS, I have not seen a *PC* that has been capapble of
averaging 30 plus frames of images in REAL TIME and display images whilst each 
frame is coming in, with resolution greater than 800x600 in monochrome (8bit+)
let alone 24bit colour!

-The main-stream multimedia maket is changing the face of image processing.
-Given this fast growing industry, prices are dropping fast. 

For niche products, that only cover standard video and not all are good
for all types of image processing, or even dealing with VCRs

-Some PC's like the IBM ThinkPad 755CD (A LAPTOP!) have video capture built in!
 
see above

> This is often fine if you only ever capture from a standard video camera, but
> the quality of image, colour depth, and other factors depend on your application
> as these types of cards are no use to connect to slowscan devices or digital
> high speed devices.
> You might have been comparing apples and bananas.

The high-volume multimedia industry will support these peripherals as well.
(It is happening already) API's will evolve to meet the demand.

No it is not as a lot of people I know have tried doing colour at true 24bit
colour and found that the quality is at best VHS which is nowhere near the 
required colour resolution. All the other devices are NOT supported and doubt 
they will be.

-> Maybe you have time to write libraries, and few costs associated with your 
-> time others don't and in a lot of establishments the cost of writing 
-> libraries can be 2-4 man weeks, which is a lot more than buying libraries.

-Which is why I hope some entrepreneur wakes up and publishes a library of
-image processing routines for the OS/2 True-color BMP and Windows True-color
-BMP file formats. (The file data can be stored in a RAM buffer and easily
-processed this way.)

I agree that something like that needs to be written and I believe that some
are working on it and I do know of some application software packages and 
libraries that are available for various platforms, OS and device support, that
do more than what you were originally comparing against. For the same or less
cost.

-Besides, I believe the time spent writing libraries from scratch using a
-32-bit OS is less costly than learning the protocols of one hardware vendor's
-system, and then being stuck with it. Especially if that system is based on
-DOS. My libraries will work on data from any multimedia capture card. Once
-I get a copy of OS/2 for the Power PC, they will really scream. A few changes,
-and they could port to Windows NT. 

Which comes back to what bus which effects the drivers and all cards are not
the same. Also drivers can vary considerably in complexity dependent on 
platform, OS, application requirements....

-If you really MUST use special, dedicated hardware for your application, then
-yes, you cannot use the techniques I suggest. But I'd say at least 98% of
-the computer vision and imaging work being done COULD be accomplished with more
-portable techniques available with 32-bit programming and multimedia support.
-This applies to both workstations and PC's.

I would say that the number the imaging cards from all manufacturers for image
processing suggest that it is larger than 2% of the market using these 
techniques. I agree that 32bit programming and linear addressing makes life more
portable and easier to optimise. Unfortunately the multimedia market while doing
some good things also tends to make those with budgets think that any video 
capture device and any camera lens will do ANY job. A bit like the scenario
I still in UK companies whereby any computer can do a processing job so the
staff doing the actual work get the worst possible computers (Xt/286) but the
more senior the manager the better the system!

I would suggest that any further comments be via EMAIL as I am sure that not
everybody wants to read all of this, or pay for the downloads.

-- 
Paul - "Any people you should meet are the products of a deranged imagination"
