Re[8]: [XviD-devel] New Motion Estimation (from sysKin) committed to branch

[Michael Niedermayer]

Hi

On Thursday 26 September 2002 03:21, peter ross wrote:
> >btw, why are there not-coded VOPs in there?
> >if they are supposed to be skiped frames then they should have a unique
> >timecode IMHO
>
> the n-vops are inserted into "empty" frames when using packed-mode.
> [foo] represents an avi frame, [-] re e.g.
>
> unpacked: [I] [P]  [B] [B] [P]  [B] [B] [P]
> packed    [I] [PB] [B] [-] [PB] [B] [-] [P]
>
> the empty frames cannot be set to length=0, because vfw decoding
> interprets this as a frame drop and doesnt bother calling the codec.
>
> so to get arround this, divx5.02 inserts a n-vop using the previous
> p-vop's timecode. 
i know, ffmpeg obviously has to handle that for divx502+ decoding, and iam not 
too happy about it as it causes problems if the decoder is feeded with raw 
m4v streams and should output the correct number of bytes consumed for 
decoding the current frame, and iam not sure if it will work with more than 1 
b frame currently

> xvid does the same to remain compatible.
i didnt know that (i only saw a few N-vops at the begin, not for every P+B 
frame
btw ffmpeg doesnt do something like that for b frame encoding

> imho, inserting a simple user_start_code would be better.
yes, than the stream would be more legel ...

[...]

Michael