[FFmpeg-devel] [FFmpeg-cvslog] avfilter: add vmafmotion filter

Michael Niedermayer michael at niedermayer.cc
Sun Oct 8 01:06:42 EEST 2017


On Sat, Oct 07, 2017 at 11:39:40AM -0400, Ronald S. Bultje wrote:
> Hi,
> 
> On Sat, Oct 7, 2017 at 6:25 AM, Michael Niedermayer <michael at niedermayer.cc>
> wrote:
> 
> > On Sat, Oct 07, 2017 at 12:12:15PM +0200, Michael Niedermayer wrote:
> > > > >>>>> On Fri, Oct 06, 2017 at 10:03:16AM -0400, Ronald S. Bultje wrote:
> > > > >>>>>> I'm open to any alternative method you suggest.
> >
>  [..]
> 
> > > About a solution, I didnt write alot about solutions as iam really fine
> > > with any solution. There are many possible solutions ...
> > > The libavfilter/af_astats.c filter i mentioned outputs statistics
> > > via metadata
> >
> [..]
> 
> > > There are other possibilities. Also Hendriks suggestion about
> > > turning metadata via some other application option into a file sounds
> > > pretty good to me
> >
> 
> psnr/ssim/vmafmotion indeed use per-frame metadata, but that unfortunately
> doesn't address the use case of using ffmpeg to quickly dump some per-frame
> stats to a file (or stdout) for quick comparison/verification (e.g. compare
> against Netflix' vmaf code reference scores) since we don't print frame
> metadata anywhere.
> 
> If someone does what Hendrik suggests (turning metadata into a file or into
> a log) then indeed there is no problem.
> 
> 
> > > or av_log as 2 alternatives to direct file write.
> >
> 
> av_log would be simpler, yes.
> 
> yet another solution (if people dislike all suggestions so far) would
> > be adding a flag to filters which are safe to use with untrusted
> > parameters. Or a flag to options which are unsafe so they are blocked
> > from being set from an untrusted input string
> >
> 
> I actually considered this myself, but didn't want to propose it because I
> personally thought it was a little hacky... I'm not against it, though.
> 
> Of these, I probably like the first one best (having some way of writing
> per-frame metadata generically to stdout or to a file, without explicit
> options), but it may also be the most complex one. Any takers to implement
> this?

if theres some consensus on what to implement, iam volunteering to
help implementing it.

(a)metadata as paul mentioned is not a complete solution as it itself
turns a filter parameter into a path to write to.

We can also leave this all as it is and just document that the filter
graph description string and some filter parameters must not be from
untrusted sources if people want to allow/support/have filters that
write to user specified pathes directly.

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

In fact, the RIAA has been known to suggest that students drop out
of college or go to community college in order to be able to afford
settlements. -- The RIAA
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 181 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20171008/e372549a/attachment.sig>


More information about the ffmpeg-devel mailing list