[MPlayer-users] pp=hb/vb/dr doesn't work while encoding?

[D Richard Felker III]

On Mon, Mar 15, 2004 at 12:27:09AM +0200, Ville Saari wrote:
> On Sun, Mar 14, 2004 at 09:20:18AM -0500, D Richard Felker III wrote:
> 
> > > I also have used deblocker after cropping by multiples of 16 and
> > > been subjectively satisfied with the results.
> ...
> > Deblock and dering filters will NEVER work after cropping when the
> > offset is not a multiple of 16, and even supporting that case would
> > require changes to the crop filter which are not likely to be added.
> > So ALWAYS apply pp first.
> 
> The result was less blocky than without the filter, so I believed it
> had worked. I used it for transcoding a certain TV-series from
> VideoCD-format to mpeg4 and now that you have told me this, I have to
> re-encode 68 episodes :-(

It will reduce the blockiness, but maybe not enough, or maybe it will
blur more than it needs to. It simply doesn't have the info to know
how strong the deblocking needs to be.

> I suggest this to be added to the manpage. I believe the same goes
> for softpulldown too. Filmdint is the only filter that the manpage
> instructs to apply as first filter. 

How about instead making the pp filter print a strong warning message
if hb/vb/dr is enabled and the quantizer table is not available. Or
maybe even an error unless the user specifies the fq (force quantizer)
option.

> > > PAL DVDs frequently have the frames-out-of-phase-problem I described in
> ...
> > You should also try filmdint with the right options, or pullup. Either
> > of these filters should repair it properly.
> 
> I have tried both with unsatisfactory results. The pullup didn't do what
> I wanted and filmdint caused audio desync.

For filmdint you'll want to specify 24:25 as your out:in field ratio.
BTW could you upload some short clips that show these filters failing?

> > > I'm still searching for a way to fix it. Median deinterlacing should
> > > theoretically work, but the field that pp=md wants to keep is just
> > > the wrong field for the material I'm trying to fix and my attempts to
> > > invert the field order have resulted either colorspace incompatibility
> > > problems or crashed mplayers :-(
> > 
> > Try vf_flip to invert the field parity.
> 
> My reference to crashed mplayers referred to just that. pp=md works fine,
> -vf flip alone also works fine, but -vf flip,pp=md,flip produces a core

Bleh, try adding scale before and/or after pp. It'll be slow but at
least it should fix the crash. G1's video filter system is buggy. :(

> dump. Colorspace incompatibility was a result of trying to
> -vf ilpack,crop=720:574:0:1,pp=md.

Yes, pp only supports planar colorspaces, not packed.

> In the light of what you said about pp, that shouldn't be possible anyways.
> However pp=md definitely has worked for me even when it has not been the
> first filter.

Deinterlacing doesn't require quantizer tables. They're just needed
for the options that clean up artifacts from quantization errors
(deblocking and deringing). Sorry I wasn't clear about the distinction
before.

Rich