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

[Corey Hickey]

Tuukka Toivonen wrote:
> [Automatic answer: RTFM (read DOCS, FAQ), also read DOCS/bugreports.html]
> On Thu, 9 Oct 2003, D Richard Felker III wrote:
> 
>>On Wed, Oct 08, 2003 at 11:36:37PM -0700, Corey Hickey wrote:
>>No problem! Thanks for performing the tests and posting your results!
> 
> 
> Damn! I missed the original post. Fortunately, I found it from the
> archives. Somebody should start collecting encoding tests into some web
> page, they're impossible to find from archives afterwards. In fact, I just
> started doing that. But please tell me if somebody already has a test page.
> 

The original authoritative page for a variety of libavcodec parameters,
written by Rémi Guyomarch, was:
http://rguyom.ath.cx/

That site seems to have been dead for a long time, so I just mirrored
the pages here:
http://fatooh.org/remitests/

Actually, as of yesterday I've started to put together a test framework
of similar scale. I have mostly adapted an earlier project of mine into
a shared-NFS-filesystem distributed testing daemon that accepts test
parameters in the form of little "ticket" files, encodes the test,
leaves the second pass in an output directory, and inserts various
results into a mysql database. What I need to do now is write a script
to generate the "ticket" files with all sorts of lavcopts combinations,
and then I'll figure out some reasonable way to pull info from the
database into some nice HTML pages.

Give me some time to do that, though... :)

> Corey, do you mind that I put you message to:
> http://www.ee.oulu.fi/~tuukkat/mplayer/tests/corey-hickey-2003-10-09.txt
> 

No problem, by all means go ahead.

> Also, those mv0 and cbp are in CVS version, right? At least I can not find
> them from mplayer-1.0pre1.
> 

Yes, that's right. If I remember correctly, Michael added them to CVS
between pre1 and pre2.

> I'm planning to do some tests myself too, from analog TV material, as soon
> as I have some free CPU cycles.
> 
> Please also benchmark next time how slow the options are.

I didn't do this earlier because I've never been too concerned about
speed. But, I know that other people are so once I get the system I 
mentioned above going, fps will be one of the results (derived from
#frames / user-time). If you want something sooner, I've attached the
script I used to perform the tests earlier. It should be fairly easy to
modify to include fps. Use /usr/bin/time -f"%U'

> Also it would be
> interesting to see rate-distortion curve (encode same video with same
> options but with different fixed quantization parameter, and plot it. It
> will show both how much an option improves PSNR at any bitrate, _or_ how
> much it decreases bitrate at given quality, which might be a better
> test...)
> 

Hmm... that should interesting... I'll probably ask you to elaborate
later on, once I get to a point where I can start adding tests and doing
new things.

-Corey
-------------- next part --------------
A non-text attachment was scrubbed...
Name: lavctests.sh
Type: application/x-sh
Size: 1836 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/mplayer-users/attachments/20031010/cf1d7069/attachment.sh>