[MPlayer-dev-eng] [PATCH] synch to new x264 revision

[Guillaume POIRIER]

Le lun 27/09/2004 à 18:26, Loren Merritt a écrit :

> > Loren, are my stats of any help?
> 
> Are those 15 encodes all you did? What are they supposed to tell us?
> I don't see any 2 encodes that differ by only 1 option, so I don't think
> one can gain anything rigorous from that test.

You're absolutely right about this... I should have set it to do more
runs, and check is before sending it to the ML.

mencgen is _supposed_ to try 15 random combinations of x264 options,
then pick the top group of options that provided best PSNR, and breed
those champions with other options.
I guess what I didn't see is that the breeding process doesn't seem to
work (but it does work with XviD and lavc). That's a bug.
I need to report that to Romain.
... add regarding what you can do out of it... you can still have a 
"manual" look at it a see what combination out of the 15 produce the 
best PSNR...


> You seem to be misinterpreting mencgen's summary of how often it 
> _tried_ each option as being some measure of what's better.

I did, I just thought the breeding process was working, and that 
the end stats were what I was looking for.
Heck, I feel stupid! :-(

I'll do some more tests as soon as the breeding is fixed.

Did you try mencgen yourself maybe?

Regards,
Guillaume