Re: [video-codec] draft-filippov-netvc-requirements-01

["Timothy B. Terriberry" <[e-mail]>]

Thomas Daede wrote:
> In addition, I think spending effort on the high latency case is not
> nearly as high a toll as you think. Pretty much the only bitstream
> feature that is needed for high latency support is bi-predicted frames.
> And Thor already has these.

Also, as I have said before, network effects dominate. Having tried this 
royalty-free codec thing several times now, I believe that to the extent 
Opus has been more successful than our prior efforts it is precisely 
because it is better on all fronts across a very wide range of 
applications. You don't build network effects by being a niche codec, 
especially when your competition (as here) is not a niche codec.

I do think focusing on interactive first has some benefits. If you focus 
on streaming/random access first, you wind up with a laundry list of 
features that are not, e.g., robust to packet loss, and thus have to be 
shut off for interactive applications with some large penalty (> 5% 
rate), while alternative features could have been designed that would 
have been nearly as good (within 1-2% rate) and also been robust to 
packet loss. This has already come up several times in Daala and we have 
opted for the latter approach. You can't do that for all features (frame 
re-ordering being one big exception, as you mention), but you can do it 
for a lot of them.