[FFmpeg-devel] [RFC PATCH] avfilter/fastdeint: import simple cpu-optimized deinterlacing algorithms from VLC

Paul B Mahol onemda at gmail.com
Sat Jan 2 17:15:16 EET 2021


On Tue, Sep 10, 2019 at 1:03 AM Aman Gupta <ffmpeg at tmm1.net> wrote:

> On Mon, Sep 9, 2019 at 3:19 PM Hendrik Leppkes <h.leppkes at gmail.com>
> wrote:
>
> > On Tue, Sep 10, 2019 at 12:00 AM Aman Gupta <ffmpeg at tmm1.net> wrote:
> > >
> > > On Mon, Sep 9, 2019 at 2:47 PM Carl Eugen Hoyos <ceffmpeg at gmail.com>
> > wrote:
> > >
> > > > Am Mo., 9. Sept. 2019 um 22:19 Uhr schrieb Aman Gupta <
> ffmpeg at tmm1.net
> > >:
> > > > >
> > > > > From: Aman Gupta <aman at tmm1.net>
> > > > >
> > > > > These are simple algorithms which can be run efficiently
> > > > > on low powered devices to produce deinteraced images.
> > > >
> > > > Please provide some numbers about the performance
> > > > (and subjective visual quality) of the new C code in
> > > > comparison to existing deinterlacers in FFmpeg.
> > > >
> > >
> > > Comparison of visual quality can be seen on VLC's website:
> > > https://wiki.videolan.org/Deinterlacing
> > >
> > > Regarding performance- none of the filters currently available in
> ffmpeg
> > > are fast enough to deinterlace video in real time on ARM chips used by
> > > popular Android or iOS devices. They're all very computationally
> > expensive,
> > > and do not have any ARM SIMD implementations. The deinterlacers from
> VLC
> > > use simple mathematical averages optimized by SIMD, and have been used
> by
> > > VLC on such devices for many years. I don't have any hard numbers to
> > share,
> > > but in my experience I can decode+deinterlace video for real time
> > playback
> > > in VLC on any cheap Android phone, whereas other ffmpeg-based players
> > > cannot.
> > >
> >
> > None of those algorithms are really worth using, none are actual
> > "deinterlacers". Blend and Mean are just plain out terrible, and the
> > other options are just dumb bob'ers which you can do with avfilter
> > as-is today with a combination of the separatefields filter (which is
> > zero-copy based on frame metadata only) and optional scaling
> > afterwards.
> >
>
> I don't disagree that many of them are overly simplistic. I only copied
> them all for completeness sake.
>
> However, as terrible as they may be they're not as bad as displaying
> interlaced frames directly. Blend and Linear produce acceptable image
> quality imho.
>
> Linear averages lines from both fields to generate a new image. Is
> something like this possible with any existing filter combined with
> separatefields?
>

I do not think its currently possible, anyway just add only linear variant
and be done.


>
> Aman
>
>
> >
> > - Hendrik
> > _______________________________________________
> > ffmpeg-devel mailing list
> > ffmpeg-devel at ffmpeg.org
> > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> >
> > To unsubscribe, visit link above, or email
> > ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".


More information about the ffmpeg-devel mailing list