[theora] OGGZ Seeking in Theora

[chris at pearce.org.nz]

 On Mon 28/06/10  7:57 PM , Stefan Weber s_weber at blyth.ch sent:
> I'd really like to implement a correct seeking functionality.

A quick hack to get visual artifact free seeking with liboggz is to seek to a time which should lie before a keyframe prior to your seek target, and then decode forwards to the target frame. This will mean seeking takes a few more MS (since you'll likely be decoding more than required), but it's better than visual artifacts after seeking, and is easy to implement to get seeking working.

We can calculate the maximum time between frames as:

max_keyframe_interval_ms = ((1 << th_info.keyframe_granule_shift) - 1) * th_info.fps_denominator / th_info.fps_numerator

So seek to max_keyframe_interval_ms before your target and pump the decode until you reach the target frame. You may need to subtract a few extra milliseconds in there if you're not using the seek-rewrite branch of liboggz to work around old edge-edge case bugs.


All the best,
Chris P.



> Whats the best method to get the time representation of a frame?
> Regards,
> Stefan
> Am 21.06.2010 11:58, schrieb Stefan Weber:    Dear all
> Conrad, thank you for your inputs and your help so far. I will use
> and test the seek-rewrite branch as soon as possible.
> I figured out that the Keyframe position in the granulepos is the
> NUMBER of the keyframe. When seeking with OGGZ, one has to supply the
> TIME-position (milliseconds) of the desired frame. So I made a short
> workaround by calculating the milliseconds position as follows:
> ms = (granulepos >> shift) * 1000 / noFPS;
> Needless to say, this won't point me to the exact keyframe
> position. Probably this rough calculation is not exact enough?
> What is the best approach to calculate the TIME-position for
> oggz-seeking out of a given frame number?
> Kind regards,
> Stefan
> 
>