[MPlayer-users] Two-passes encoding - What happens exactly?

[Alain Barthélemy]

Le Thursday 03 June 2004, 19:43:38 ou environ D Richard Felker III <dalias at aerifal.cx> a écrit:
> On Fri, Jun 04, 2004 at 12:02:39AM +0200, Alain Barthélemy wrote:
> > Le jeudi 03 juin 2004, 11:54:19 ou environ D Richard Felker III <dalias at aerifal.cx> a écrit:
> > > On Thu, Jun 03, 2004 at 09:36:00AM +0200, Alain Barthélemy wrote:
> > > > > Is that the highest resolution your tv card is capable of? I use 
> > > > > "width=720:height=576".
> > > > 
> > > > Normally it should be 640x480 but I made lately a quick test with resolution
> > > > 720x576 and it worked.
> > > 
> > > It depends on whether you have NTSC or PAL.
> > 
> > I have PAL thus it should be 640x480. Thanks for reminding me.
> 
> Backwards. PAL is 576 lines, NTSC is 480. You're free to use whatever
> horizontal resolution you want with either, but if you want square
> pixels it's 640 for NTSC and 768 for PAL.

Thanks. Here I really misinterpreted everything. I'll go back to school. I
found an article (you must know it I suppose because it starts with
www.mplayerhq.hu)on the Net that talks (in French) about frames, fields,
telecine, etc ... I did not read it before because there is non telecine with
PAL.


#############################

> 
> > > -lavcopts acodec and abitrate are useless unless you're using -oac
> > > lavc. For -oac mp3lame you use -lameopts!
> > 
> > Useless? OK! But you don't enligten me. I must understand that -lavcopts acodec
> > and abitrate are *only* useful if I have -oac mp3lame lameopts cbr:br=96. *But*
> > does not it do the same as -ovc lavc -lavcopts acodec=mp3:abitrate=mp3. Should
> > not I use one or *exclusive* the other option?
> 
> You still misunderstand. Here:
> 
> Use -lavcopts acodec=???:abitrate=??? if you will be using lavc to
> encode audio, i.e. -oac lavc.
> 
> Use -lameopts ... if you will be using lame to encode audio, i.e. -oac
> mp3lame.
> 
> There's no point in using both, since you'll be configuring a module
> you're not going to use!


Not really misunderstood. That's why I wrote *exclusive* OR but it is true I
confused ovc lavc and oac lavc. Now I'll stick to -oac mp3lame -lameopts
-cbr:br=***.

###########################


> 
> > > > > "-vf denoise3d=5:4:7" should give better quality than "-vf pp=tn" (But 
> > > > > needs more CPU-time).
> > > > 
> > > > I'm not familiar with denoise. Still learning about video compression (and still dumb).
> > > 
> > > Well don't use pp=tn. It removes noise but ADDS blocking artifacts
> > > which are much worse!

I have two advices now. Thanks to both. pp=dn is thus not perfect. Does it mean
that to reduce image noise -vf denoise3d=5:4:7 is better to use. I have read in
man pages that the default parameters for denoise3d are 4:3:6. I am no
specialist to know if 5:4:7 is better or not. It is a question of trial of
course.  Of course I'll have to deinterlace, thus: -vf crop=*:*:*:,denoise3d=5:4:7,pp=lb,scale=*:*

############################
> > 
> > Thus I keep now:
> > 
> > mencoder -tv driver=v4l:width=640:height=480 -ovc lavc -lavcopts
> > vcodec=mpeg4:vbitrate=1800:mbd=2:v4mv -oac mp3lame -lameopts cbr:br=96 -vf
> > crop=620:460,denoise3d=5:4:7,scale -zoom -xy 640 -sws 1 movie.avi
> > 
> > ... till someone tells me the image will be completely distorted.
> 
> It will, unless your video miraculously happens to be noninterlaced
> (actually 2:2 pulldown), which I expect was the case for your example.
> You have to use a lot of care to scale interlaced video, or
> deinterlace it first. I would recommend against upscaling in any case.
> And be sure your final sizes are multiples of 16, etc. Also use the
> right capture resolution for PAL, ...

Apparently the result is good. It was "Longest Day" in B/W. In fact it is send
by FR3 in SECAM format but I receive it with cable in PAL format thus it was
converted by the cable company and I don't have the technical details of the
conversion.

Thanks for all that help.

> 
> Rich

-- 
Alain Barthélemy
cassandre at bartydeux.be
http://bartydeux.be
Linux User #315631