[MPlayer-users] PSNR tests: mbd=2, mv0, trell, and cbp (and...

Michal Seliga michal.seliga at visicom.sk
Mon Oct 13 10:47:08 CEST 2003


> Date: Mon, 13 Oct 2003 10:43:16 +0300 (EEST)
> From: Tuukka Toivonen <tuukkat at ee.oulu.fi>
> To: Corey Hickey <bugfood-ml at fatooh.org>
> Cc: "MPlayer user's list." <mplayer-users at mplayerhq.hu>
> Cc: ffmpeg-devel at lists.sourceforge.net
> Subject: Re: [MPlayer-users] PSNR tests: mbd=2, mv0, trell, and cbp (and
>  other stuff)
> Message-ID: <Pine.GSO.4.58.0310131028180.24728 at stekt37>
> In-Reply-To: <3F8706B9.6020004 at fatooh.org>
> References: <3F8501F5.5050205 at fatooh.org>
>  <20031009210242.GZ2856 at brightrain.aerifal.cx>
>  <3F8706B9.6020004 at fatooh.org>
> Content-Type: TEXT/PLAIN; charset=US-ASCII
> MIME-Version: 1.0
> Precedence: list
> Reply-To: "MPlayer user's list." <mplayer-users at mplayerhq.hu>
> Message: 6
> 
> On Fri, 10 Oct 2003, Corey Hickey wrote:
> 
> >[Automatic answer: RTFM (read DOCS, FAQ), also read DOCS/bugreports.html]
> 
> Looks like Pine doesn't allow me to quote more than this one line...
> anyway...
> 
> I made some tests with mencoder/libavcodec using different set of
> options and the results are here:
> ....I cut the rest....

firstly, good work, I wanted to make something similar long ago, but never had a time to finish it...

from my tests I  got these results I thought that may be interesting for you...

1)  if video is noisy then adding B frames always make result look worse (and with constant vqscale the result is also bigger).
     if source si in good quality (dvd) they always help. people in forum say that b frames always lower result video quality,
     anyway, I never noticed it (yes, psnr is a bit worse). anyway with b frames video looks worse while moving in it (rewind etc.).

     what I did was that I tried to encode video in all possible b-frame variants (0,1,2,3,4) with constant vqscale=2, and for final packing used the     
     shortest version. for me it made the best results, but in forum someone wrote that it wasn't good idea :-( so I don't know, maybe I just don't
     see what others do... for now I better don't use b-frames at all (there was problem that divx codec doesn't play these files well, and it is
     boring to always persuade people to use ffdshow...)

2)  yes, qpel makes bigger files... but not everytime, with dvd source it is sometimes a bit better. maybe it depends on quality of source again

3)  I wouldn't use video from tv for tests if I were you (some part of hq dvd movie would be better). At least I had problems with artifacts and that
     I cannot grab tv directly to divx, So I have to use mjpeg, and mjpeg creates some more artifacts...denoise helps a bit, but they are still visible
    and divx result looks much worse (encoded dvd movie usually looks quite well with bitrate 1000, same resolution tv looks bad even with 2000 :-(
    otherwise all artifacts were blurred. try to make your tests (at least with b-frames and qpel) with good source (ripped dvd) and maybe the results
    will differ a bit.

anyway, I am glad that now I don't have to do tests myself ;-)



More information about the MPlayer-users mailing list