[foms] WebM Manifest

Timothy B. Terriberry tterribe at xiph.org
Thu Mar 17 14:53:50 PDT 2011

> Yes, that's what I mean. Statistically I think there are good chances
> the different variants/bitrates passed through the same encoder (and
> same "GOP size") should produce the same keyframe positions with
> little differences. Not having to synchronize the encoders would make

Keyframes may "often" be in the same place, but "often" starts dropping 
off pretty fast as you add more bitrates (90% drops to 66% when you add 
four more versions). This is something that's better controlled in the 
encoder, as Mark Watson suggested. libtheora, for example, will place 
keyframes in the same position regardless of rate if you use the same 
first-pass statistics for different second-pass encodes. It makes no 
guarantees otherwise (and Greg Maxwell wanted to break even this 
behavior to get better compression... but he never sent me patches).

> life of the encoder (and likely the rate control process) easier. And
> on the playback side there's hardly any drawback. But that means
> players have to handle the case they want to switch but the next
> keyframe is not available in the variant it wants to switch to.

This also assumes this is even expressible in the manifest. I don't know 
DASH well enough yet to say if it is. And it leaves you with the 
possibility of switching to a rate and not being able to switch out of 
it for a long time. Imagine a relatively still scene with only forced 
keyframes at periodic intervals: if these are misaligned at the start, 
it could be many tens of seconds before there is another opportunity to 
switch. This sort of thing is not as uncommon as you think.

More information about the foms mailing list