[vorbis-dev] video codec

[Gregory Maxwell]

On Sat, Feb 10, 2001 at 12:03:34AM +0800, Steve Underwood wrote:
[snip]
> Your useless data may not be mine. Years of DSP have taught me that I
> can predict only 80-90% of the outcome. The other 10-20% can be a
> killer! Don't wait too long to find out.
[snip]

Obviously, you don't want to ignore any problem set.. However, the current
codec differs from all other production codecs in one signifant way: Use of
a 3d transform insted of motion vector compensation. 

I think that whole new kind of knoweldge will need to be created for this to
work well, and it will be some time before we get to boring details.

For example, in my testing with 3d transforms, I found temporal ghosting to
be a big problem at sudden scene changes (I suspect because we have so
little temporal resolution compaired to spacial resolution and the
thresholding doesn't concentrate power in depth-space sufficently to cope
with the low resolution these). Perhaps we need to generalize SET
to 3d so we can clamp the depth of a transform to non-power of two sizes
with frame level granularity, then detect scene changes and chop the
transform.

--- >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.