[FFmpeg-devel] [RFC] better ogg seeking, keyframes only

[Måns Rullgård]

Diego Biurrun <diego at biurrun.de> writes:

> On Thu, Jun 26, 2008 at 12:35:45PM +0200, Michael Niedermayer wrote:
>> On Sun, Mar 23, 2008 at 04:37:05PM +0100, Diego Biurrun wrote:
>> > On Wed, Mar 19, 2008 at 01:16:16PM +0100, Reimar D?ffinger wrote:
>> > > I admit this is a rather ugly patch, but it improves things a lot, so
>> > > maybe you can have a look and get it committable.
>> > > There are several issues with seeking: for audio, PKT_FLAG_KEY is never
>> > > set, so the patch ignores it there, the data_offset set by the code in
>> > > util.c is wrong, so this code overrides it, and the most annoying
>> > > thing: a keyframe most likely is split over multiple ogg pages, so
>> > > this code just returns the start of the last page read by the last
>> > > get_packet as start offset for the keyframe - this means we might
>> > > actually end up a few frames before the key frame (in theory up to 255),
>> > > if the encoded frames are very small.
>> > 
>> > Hmm, I still get artifacts when seeking in Theora video, for example
>> > with the samples from mphq...
>> 
>> I can confirm this, it doesnt work here either
>
> So who knows how to fix this? :)

I do.  Can you please look up the home addresses of the Xiph devs,
while I fetch the shotgun?

-- 
M?ns Rullg?rd
mans at mansr.com