codec optimization

[Albert Cahalan]

On 2/26/07, Chih-Chung Chang <jochang at gmail.com> wrote:
> On 2/26/07, Albert Cahalan <acahalan at gmail.com> wrote:
> >
> > (BTW, I can imagine a fine low-CPU codec. You can use
> > 100% P-frames w/o motion vectors if the decompressor
> > is carefully specified regarding rounding. Start the stream
> > with an implied solid-color frame, probably grey. Cuts and
> > joins can be done anywhere at the cost of one or two
> > recompressed frames.)
>
> I think this can be done by using the INTER_NOMV mode in Theora.

That sounds like I-frames, not P-frames. P-frames depend on
preceding frames. Normally you can't keep going with them
forever because round-off errors in the decoder will accumulate.
If the decoder is more standardized though, the encoder could
determine exactly what the decoder is using as a reference.

So INTER_NOMV won't work. Theora also has the space-filling
curve that was a neat but bad idea.

If being XO-specific is OK, one could even plan for video to be
played back without the swizzle. By using 12x12 blocks
(yes, you can do a transform on non-power-of-two -- see FFTW)
and a quantization table that makes special allowance for the
red/green/blue stripes, you could get color video while only
decoding a single channel of data.