[vorbis-dev] Tarkin at last

[Erik J. Kruus]

Jack Moffitt wrote:

> the quality levels we worked with were 95% discard.  even 99% discard
> was pretty decent.
>
> zero-tree packing was what monty and i discussed to reduce the size, and
> hatamoto, monty, and i had some ideas on packing colors efficiently.
>
> obviously there are a lot of easy improvements that can be made.
>
> 1) find a better wavelet
> 2) zero-tree packing
> 3) color packing
>
> any other ideas?
>

omething I'ld like to call an adaptive wavelet, but really isn't a wavelet at all,

except for falling into the predict&correct category, based
on the concept of edge-directed interpolation
    http://www.ee.princeton.edu/~lixin  (download the thesis)

The bad and a bit of good:

I think his implementation can be improved fairly easily.  I don't know if it's
patented or how. He only treats still images/keyframes, but the idea
extends to 3D trivially (and would do things like slow motion reasonably well, I
imagine
-much- better than a classical wavelet analysis). Straightforward implementation,
even in 2D is
pretty slow (should be done via SVD I think, and ARPACK[++] (free) is my
recommendation
for that).  It's too much work.  (It's easier just to live within the classical
wavelet world).

Erik.

>
> jack.
>
> On Thu, Jan 04, 2001 at 12:30:51PM -0500, Gregory Maxwell wrote:
> > On Thu, Jan 04, 2001 at 11:09:37AM -0600, Aaron Plattner wrote:
> > > 95 seems way too high.  Try ./tarkin 20 or ./tarkin 10 (quality 1 is pretty
> > > impressive for being 128k).  20% seems to make output.tark be the same size
> > > as all the .pgm's combined.  (Of course, I don't actually know how it works
> > > yet, so take that with a big grain of salt).
> >
> > Esp considering it's just performing thresholding, no zero-tree packing, no
> > pre/postprocessing of the input (it should be degammad mapped into log
> > space), and the current packing is expansive.
> >
> > Anyone know if there is a zero-tree packing implimentation that is
> > generalized to n-space?
> >
> >
> > --- >8 ----
> > List archives:  http://www.xiph.org/archives/
> > Ogg project homepage: http://www.xiph.org/ogg/
> > To unsubscribe from this list, send a message to 'vorbis-dev-request at xiph.org'
> > containing only the word 'unsubscribe' in the body.  No subject is needed.
> > Unsubscribe messages sent to the list will be ignored/filtered.
>
> --- >8 ----
> List archives:  http://www.xiph.org/archives/
> Ogg project homepage: http://www.xiph.org/ogg/
> To unsubscribe from this list, send a message to 'vorbis-dev-request at xiph.org'
> containing only the word 'unsubscribe' in the body.  No subject is needed.
> Unsubscribe messages sent to the list will be ignored/filtered.

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'vorbis-dev-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.