P3D 3D impressions from SID 98 (1 of 3)

[roberts@xxxxxxxxxxxxxxxxx (John W Roberts)]

Here are some of my notes from SID '98 (no doubt others will be coming
in as well). These are only my impressions of some 3D related items -
definitely no connection to any views of my employer. For further information,
contact SID and buy the relevant documents. Any errors are presumably
misinterpretations or typos on my part. Please don't annoy the speakers
with juvenile comments based on my impressions - make a serious study of
the original documents before contacting them. [In other words, don't judge
anybody based on my impressions - find out for yourself.]

Seminar M-7: Multimedia Accelerators 
 - The complexity of MAs is increasing faster than that of CPUs -
   current devices can have 3.5 million transistors, and 7 million
   transistor devices will be out soon. The crossover point where
   the complexity of MAs exceeds that of desktop CPUs is expected to
   be sometime in 1998. At present, performance is increasing by a
   factor of two about every six months. And yet the price point for
   MAs is much lower than that of CPUs - the 3.5 million transistor
   model sold for around $30.
 - The speaker suggested that the marginal benefit per dollar spent
   is much greater for MAs than for CPUs - including a $30 MA rather
   than a $20 one provides far greater benefit than spending $10 more
   on the CPU.
 - The most dynamic area in MA design is 3D graphics functions - while
   many different techniques are involved, the general idea is that
   the MA actually creates a three-dimensional "universe", with three
   dimensional objects in it, then maps this onto a 2D display output.
   In principle, left and right views could be tapped off from this
   three dimensional model. (I asked how much of a market it would take
   for them to add a stereo sync output to their chips - the answer
   (speaker's personal opinion) was "100 million units", but he added
   that other companies might want to go after this market - there are
   about 40 companies in the MA business.)
 - Objects in the 3D model are built up of polygons (in practice, triangles).
   Current MAs can possibly handle up to several million triangles
   per second - but the CPUs can't handle their part of the job beyond
   about a million triangles per second.
 - Texture mapping is very important for advanced 3D functions - a relatively
   small number of large triangles defines the shape of an object, then
   a finely detailed image is mapped onto the surface of that object.
   [This has been discussed on P3D - for example when one image of a
   stereo pair is out of focus, a 3D image can often be viewed in this
   way]. Texture mapping saves a tremendous amount of computation compared
   to the brute force method of defining a textured surface with millions
   of triangles, thus allowing multiple objects to be handled in real time.

M-10: HDTV Status and Prospects
 - Broadcast HDTV uses extreme MPEG-2 compression to fit its signal
   within the allocated 6 MHz bandwidth.
 - MPEG-2 compression makes extensive use of motion prediction/interpolation
   between frames. Because of this, rapid changes in the image severely
   tax the ability of the system to accurately reproduce the appearance
   of the source. The rapid cuts in "MTV sequences" are one of the ways
   used to test the limits of the compression methods. One of the worst
   natural sequences to reproduce is lighting flashes - the scene remains
   essentially unchanged, but the luminance varies rapidly, which drives
   motion detection algorithms crazy.

SID Luncheon Address: "Digital Photography: A Landscape Photographer in
   the Digital Age", by Stephen Johnson of Stephen Johnson Photography
   in Pacifica, CA:
 - The speaker is a professional landscape photographer who now uses
   digital exclusively. He spoke highly of digital photography, and
   criticized the limitations that traditional film imposes on the
   reproduction of reality.
 - He uses a digital scanning insert in a conventional 4x5 camera body.
   Resolution is 6000x7520 pixels, and takes about 2.5 minutes (or is that
   1 minute?) to acquire. Files are ~132 MB, which expand to ~170 MB
   in the format for the printer. Printing is done on a direct
   pixel-for-pixel basis. (The prints on display appeared to be about
   25-30 inches diagonal, and were phenomenal.)
 - Natural scenes include a lot of pastel colors - the most widely used
   film tends to produce highly saturated colors - digital photography
   can be used to reproduce the colors as they were in the original.
 - With normal exposure, film tends to produce black shadows because of
   limited dynamic range (perhaps 7 stops at best). The digital camera the
   photographer uses has about 11 stops of dynamic range - he can pick up
   detail in the shadowed sides of  burned trees, photographed against
   the sky in bright sunlight.
 - The speaker criticized the way film photography has come to define
   how we think the world looks - he has seen people disappointed by
   the real Grand Canyon because it "didn't look like the postcard".
   Digital photography offers a chance to see the world from a perspective
   other than film.
 - The camera can be mounted on a rotating drive, and used to record
   360 degree panoramics - this can also be done in stereo. (He showed
   a demo of a 360 degree stereo room interior.)

[Continued]


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