Re[6]: [XviD-devel] New Motion Estimation (from sysKin) committed to branch

[Christoph Lampert]

On Wed, 25 Sep 2002, Radoslaw 'sysKin' Czyz wrote:
>
> Could someone PLEASE commit the files to cvs? I admit, I'll just have
> to learn how to do it myself. But not now ;)



Yes, could somebody (else) please do? 
I won't be home before tomorrow or later. 


 
> Ah, and there is another thing - new api should have a flag telling if
> we want to use P/B decision or not (i.e. have fixed number of
> b-frames).

e.g.    XVID_DYNAMICBFRAMES  
or      XVID_BFRAMESDYNAMIC 

And I guess max_bframes gives the maximum number (so it should be large in
dynamic mode). Please note that the distance between two non-bframes must
never be more than 1 second... timecode couldn't handle that at the
moment. 


> In fact, P/B decision is a complex thing and maybe other people will
> write different procedures for it, so maybe we should have a number
> (or more flags?) to allow choosing between different procedures.
> Same for motion estimation itself. What do you think?

Yes, P/B decision is complex, but we don't really need more than one
possible way. It's "survival of the fittest". 
It's great that we have one now.  If another one comes along and it's
better than the old one, we should replace the old one by the new. 


For ME it's slightly different, because ME really is a key factor of
encoding and different MEs have different advantages. 

Current is very good, of course, but we might want to add a simpler/faster
one (if this is possible at all) and/or something like FullSearch for
reference encodings. 



> Ah, and one more thing. I don't have much (any?) reports telling if
> B-frames work or not. Edouard Gomez seems to have HUGE problems with
> it, but I don't know why.
> @Koepi - please tell me if the file you've encoded decodes fine with
> xvid (by Vdub). That's important because both ffdshow and divx5
> decoders have problems with variable number of Bframes.
> There are two ways - either it will decode, or will show artifacts and
> crash.
> @everyone: please do similar tests... just a short clip, divx5
> compatibility mode, quant ratio of 150%, max bframes = 4 (but only
> with the pach I attach here).
> I can provide with VfW .dll if anyone wants - wait, I'll upload it at
> this very moment: http://users.bigpond.com/syskin/xvid7.dll
> Happy testing!