Accelerated MPEG Compression of Dynamic Polygonal Scenes

Credits

For those curious how things were implemented, here's a full list of tools and things used:

User interface
Tcl/Tk, by John Ousterhout <ouster@sprite.berkeley.edu>
glxwin widget, by Michael Halle <mhalle@media.mit.edu>
glxaux utils, by Dan Wallach <dwallach@cs.princeton.edu>
MPEG Encoder
mpeg_encode, by Larry A. Rowe, Kevin Gong, Ketan Patel, and Dan Wallach, anonymous ftp (Click here for the MPEG directory)
The optical flow chart (figure 3) was generated with Iris Explorer.

The MPEG frame dependency chart (figure 4) was drawn with xfig.

The graphs were generated with GnuPlot.

Perl, by Larry Wall, was used extensively to massage the MPEG statistics into a format GnuPlot could digest.

The paper was written in LaTeX -- an entirely unpleasant experience.

The slides shown at SIGGRAPH were generated by Microsoft PowerPoint 3.0 (another unpleasant experience) with a number of figured generated by Iris Showcase and the NetPBM utilities. Far too much of the PostScript had to be edited by hand because either Showcase or PowerPoint didn't do the right thing.

The texture map on the pyramids is an Alaskan totem-pole, photographed by Philip Greenspun and used with permission. Check out his book, Travels with Samantha.

Everything (except the MPEG compression itself) runs in real-time on a Silicon Graphics R4000 Crimson RealityEngine. All the compression tests were run on 352x240 pixel images, although the color figures here are at higher resolution, so you can more easily see what's going on.

Various thanks:

An extra special thanks goes to David Laur for fighting with our film recorder to insure the pictures in the Proceedings weren't too dark or too red or too washed out.

Thanks also to the referees who asked lots of insightful questions, leading to numerous improvements in the "Rendering the Optical Flow" section.


Dan Wallach, CS Department, Princeton University