[FFmpeg-devel] Cooperative coding between (N)LE and FFmpeg; possibly quality improvements?

Stefan de Konink stefan at konink.de
Sat May 7 22:17:03 CEST 2011


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA512

Op 07-05-11 21:42, compn schreef:
> On Sat, 07 May 2011 21:21:13 +0200, Stefan de Konink wrote:
>> What if there was some sort of a 'hint' presented by an editor; make the
>> next frame an intra-frame. Or using the knowledge of spatial positions
>> layers in the editor, or their movement to express prediction.
> 
> so you want the ability to specify the frame type yourself in some kind
> of playlist ala:
> 
> frame000001 I
> frame000002 P
> frame000003 B
> frame000004 P
> frame000005 B
> ...
> ?

Not quite, given that the NLE knows the 'raw' footage is non-continous,
it could inform the encoder; 'totally new scene upfront' - thus I-frame.
This might reduce artifacts. I think going into P and B should be an
encoder decision. But for example a moving layer in a certain direction
might be a good motion prediction hint.

My inspiration came from Eurosport recently where there were so many
artifacts in the Snooker championship at scene switches (MPEG2 playout).
Even for a digital videomixer these hints could be interesting.


> are there programs that give the user that much power ? it seems like
> each user would have to know the intricate details of each codec before
> encoding any video.

Not that I am aware of. But there are a group of people now implementing
'Lumiere'. Getting some fundamental things in to a new NLE to optimise
the encoding of the video might not be such bad idea to design from. It
is just forcing the encoder in some direction - which may or may not be
optimal...


> not that i am for or against your idea, i am just curious.

I wonder if with what you suggested, the frame by frame hints, some
benchmark can be done 'can an external program with full knowledge
outperform the scene detector'. I don't know the outcome, if the outcome
is: a general solution works better, even for the specific case, so be it.


Stefan
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.17 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEAREKAAYFAk3FqL8ACgkQYH1+F2Rqwn1uUwCeIFX1loWhwy/WoD/9Xv85Ddwk
puUAnR3cUHknYpZ130yCzrxgh/CF6CcI
=+2nf
-----END PGP SIGNATURE-----


More information about the ffmpeg-devel mailing list