[MPlayer-dev-eng] 3 pass encoding syntax poll

[Guillaume POIRIER]

Hi,

Le samedi 05 février 2005 à 03:26 -0800, Loren Merritt a écrit : 
> When I came up with the current syntax for N-pass encoding with lavc or 
> x264 (pass=1 on the first pass, pass=3 on all subsequent), I though it was 
> the most logic method: N-pass simply enables both 1st pass stats file and 
> 2nd pass ratecontrol, so I used 3 == 1|2.
> However, I have seen many confused users, so apparently this isn't so 
> intuitive, and nobody reads the section in the manpage describing its 
> usage.

That's really a pity. But I have to admit that unfortunately sometimes
the very doc does confuse users when it discusses "3-pass encode" which
sometimes means "1:audio, 2:1st pass video, 3:2nd pass video".

So I guess you're right about the syntax being somewhat confusing.


> So I have decided to change the syntax and/or behavior of multipass 
> encoding. But since I was wrong the first time, I will now ask opinions 
> before choosing a replacement.
> 
> Currently:
>    pass=2 performs 2-pass encoding, and can't be part of N-pass.
>    pass=3 is used on all passes beyond the 1st of a N-pass encode.
> 
> Possibilities:
> * Specify both pass number and total number of passes: "pass=1/3".
>    (Of course, default to 2-pass encoding if denominator isn't specified.)
>    Benefit: allows 3-pass-aware turbo.
> 
> * Change "3" to "N". Disadvantage: I fear people will still try to use
>    "pass=2" in the 2nd of 3.
> 
> * Separate boolean option "npass".
> 
> * Always perform N-pass encoding. (It is, after all, identical to 2-pass
>    if you run only one Nth pass.)

I do think the last option is the very best. I mean: whenever you can do
some foulproof thing, I'd say: Just Do It!
We, geeks, may love to be able to tweak whatever a crazy (but smart)
mind created, but our Joe-six-pack really doesn't give a damn.
But he will like if we say: "Well, if you want to improve your encode,
just run it one more time, whatever your settings were".


> Additionally (especially in the last option), I think it would be 
> beneficial to use separate statsfiles for each pass, so that cancelling an 
> encode doesn't erase the results of previous passes.

I buy that.


> This would require 
> changing the -passlogfile option somehow, since there would be multiple 
> files to specify. How best to go about this?

Just out of my competence I guess...
Could the "subopt" parser help here? How about something like:
-passlogfile pass1=filename,pass2=filename

Regards,
Guillaume