[photo-3d] 2D-3D

[Ron Beck <rbeck@xxxxxxxx>]

First, if I'm not mistaken, the Toy Story characters were actually drawn
in 3D using x, y and z coordinates and could be rotated in any
direction.

Second, Toy Story is an old movie.  I'd be more curious about the
statistics for Toy Story 2 as processing power and storage capabilities
have increased quite a bit since the original came out.  I can buy a
Terebyte of Raid 5 redundant storage for $85,000.00.  Not exactly an "at
home" price, but certainly much less than it cost a couple of years ago.

Third, there's a difference between creating 3D characters (true 3D
coordinates) such as the Toy Story characters and extrapolating the 3D
from a 2D source such as making a Bugs Bunny VM reel.

Lastly, I can concieve of a program that, via a set of rules,
automatically create a 3D image from a 2D image.  For an extremely
simple example consider the following, if you have a person in front of
a building, extrapolate the distance the person is in front by allowing
that a person is, on average, 6ft tall and the floors of a building are,
on average, 10 ft tall.  If the person in front of the building is about
the height of a single window, they're close to the building.  If the
person appears to be more than two windows tall, they're further away
from the building.

I think of the process as an Artificial Intellegence type of system.  As
you go through conversions and define rules, the process can become more
automated.  Certainly there are those who have been doing 2D-3D
conversions for years and with the current compute power, it would be
great to try to harness their knowledge into a series of rules.

The other interesting factor in this is that motion pictures often do
contain additional 3D cues on subsequent frames of the film.  This
information could also be used to assist in rendering 3D models from 2D
information.  Besides, Ted Turner doesn't have a bunch of people
colorizing each frame of all the movies they colorized.  They have a
computer that does most of the work.  Someone identifies subjects and
the colors they should have and the computer updates the frames that the
subject is in.  Is there labor involved? certainly.  Do they have to
physically color each frame? I don't think so.  I see 2D-3D conversions
following the same path.

Besides, never say never.  IMHO, everytime someone says "it can't be
done", they're proven wrong by someone who's doing it.

Regards,
Ron

bmaxey1@xxxxxxxx wrote:
> To give you an idea of what the current state of the art is, consider Toy
> Story:
> 
> ** 10 separate texture maps to get the skin to look right. Still does not
> come close if you are to duplicate nature with exactness.
> ** Machine hours needed to render was over 800,000
> ** Maximum weekly output: 3.5 minutes of completed film.
> ** Total storage required for all final frames: 500 gigabytes
> ** Total storage required for all film information: 1 terabyte
> ** Rendering an individual frame: Up to 20 hours
> ** 110 computers operating on a 24-hour basis
> 
>  I only bore the list with the above because you most likely have all
> seen TS, and it gives you a reference to see what might need to be done
> in the way of computer power if natural looking 2D to 3D conversions are
> to be done.
> 
> Bob the Crackpot
> 
> [Non-text portions of this message have been removed]