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

[Ronald S. Bultje]

Hi,

On Sat, Sep 30, 2017 at 3:40 PM, Carl Eugen Hoyos <ceffmpeg at gmail.com>
wrote:

> 2017-09-30 20:30 GMT+02:00 Ronald S. Bultje <rsbultje at gmail.com>:
> > Hi Carl,
> >
> > On Sat, Sep 30, 2017 at 2:19 PM, Carl Eugen Hoyos <ceffmpeg at gmail.com>
> > wrote:
> >
> >> 2017-09-30 19:47 GMT+02:00 Ronald S. Bultje <rsbultje at gmail.com>:
> >> > Hi Carl,
> >> >
> >> > On Sat, Sep 30, 2017 at 1:31 PM, Carl Eugen Hoyos <ceffmpeg at gmail.com
> >
> >> > wrote:
> >> >
> >> >> Hi!
> >> >>
> >> >> 2017-09-15 22:47 GMT+02:00 Ashish Pratap Singh <ashk43712 at gmail.com
> >:
> >> >>
> >> >> > +static int query_formats(AVFilterContext *ctx)
> >> >> > +{
> >> >> > +    static const enum AVPixelFormat pix_fmts[] = {
> >> >> > +        AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV422P,
> >> >> > AV_PIX_FMT_YUV420P,
> >> >> > +        AV_PIX_FMT_YUV444P10, AV_PIX_FMT_YUV422P10,
> >> >> > AV_PIX_FMT_YUV420P10,
> >> >>
> >> >> Is the algorithm only defined for these formats and bit-depth
> >> >> or are there just missing features?
> >> >> Gray and gray10 come to mind...
> >> >>
> >> >
> >> > Great question! I _believe_ that vmaf overall is luma-only, so it
> should
> >> be
> >> > entirely independent of chroma.
> >>
> >> Then imo, above function is just wrong, it should check for
> >> non-rgb or similar (think of YUVA444 and friends).
> >>
> >
> > I don't think I'm familiar enough with lavfi to send a patch, can you
> send
> > one? What I've asked Ashish to do (and what he's done here) is simply to
> > reproduce as closely as possible what Netflix' code does, and they only
> > support 420, 422 and 444 for 8 and 10 bits/component. I'm happy to
> support
> > more if I know how to.
>
> Attached patch also support GBRP, I don't know if this is a good or
> bad idea.
>

I would personally probably err on the side of caution, but no strong
opinions. If this works (I assume it does), I guess it's fine (?).

Ronald