[daala] DAALA codec efficiency
jack at metajack.im
Sat Oct 12 21:00:48 PDT 2013
> The only useful comparisons are between iterations of the codec itself,
> and even those have different goals. I'm sure one of the developers
> would have more insight into this, but to say comparing to other codecs
> is premature at this state is an understatement.
One reason for this is we're still figuring out how everything will
work and fit together. For example, we have to do intra-prediction in
the frequency domain because of the lapped transforms. We're spending
a lot of time exploring and validating these ideas right now to make
sure they will work and perform as expected.
We haven't been spending much time on compact representation since
we're switching out major pieces still, and for example, the main
quanitzation stuff, PVQ, is still under heavy development and accounts
for almost all the bits. It's going to be hard to compare Daala
directly to other codecs for a while yet.
The intent is to outperform those codecs you tested against. It will
take some time.
More information about the daala