[FFmpeg-devel] [PATCH v3] avutil/pixfmt: improve definition of AVColorRange

[Jan Ekström]

On Sun, Sep 20, 2020 at 9:34 PM Vittorio Giovara
<vittorio.giovara at gmail.com> wrote:
>
> On Sun, Sep 20, 2020 at 7:12 PM Jan Ekström <jeebjp at gmail.com> wrote:
>
> > As it was brought up that the current documentation leaves things
> > as specific to YCbCr only, ICtCp and RGB are now mentioned.
> > Additionally, the specifications on which these definitions of
> > narrow and full range are defined are mentioned.
> >
> > This way, the documentation of AVColorRange should now match how
> > most people seem to read interpret it at this point, and thus
> > flagging RGB AVFrames as full range is valid not only according to
> > common sense, but also the enum definition.
> > ---
> >  libavutil/pixfmt.h | 54 +++++++++++++++++++++++++++++++++++++++++++---
> >  1 file changed, 51 insertions(+), 3 deletions(-)
> >
> > diff --git a/libavutil/pixfmt.h b/libavutil/pixfmt.h
> > index a46acf3c5e..46ef211add 100644
> > --- a/libavutil/pixfmt.h
> > +++ b/libavutil/pixfmt.h
> > @@ -530,12 +530,60 @@ enum AVColorSpace {
> >  };
> >
> >  /**
> > - * MPEG vs JPEG YUV range.
> > + * Visual content value range.
> > + *
> > + * These values are based on definitions that can be found in multiple
> > + * specifications, such as ITU-T BT.709 (3.4 - Quantization of RGB,
> > luminance
> > + * and colour-difference signals), ITU-T BT.2020 (Table 5 - Digital
> > + * Representation) as well as ITU-T BT.2100 (Table 9 - Digital 10- and
> > 12-bit
> > + * integer representation). At the time of writing, the BT.2100 one is
> > + * recommended, as it also defines the full range representation.
> > + *
> > + * Common definitions:
> > + *   - For RGB and luminance planes such as Y in YCbCr and I in ICtCp,
> > + *     'E' is the original value in range of 0.0 to 1.0.
> > + *   - For chrominance planes such as Cb,Cr and Ct,Cp, 'E' is the original
> > + *     value in range of -0.5 to 0.5.
> > + *   - 'n' is the output bit depth.
> > + *   - For additional definitions such as rounding and clipping to valid n
> > + *     bit unsigned integer range, please refer to BT.2100 (Table 9).
> >   */
> >  enum AVColorRange {
> >      AVCOL_RANGE_UNSPECIFIED = 0,
> > -    AVCOL_RANGE_MPEG        = 1, ///< the normal 219*2^(n-8) "MPEG" YUV
> > ranges
> > -    AVCOL_RANGE_JPEG        = 2, ///< the normal     2^n-1   "JPEG" YUV
> > ranges
> > +
> > +    /**
> > +     * Narrow or limited range content.
> > +     *
> > +     * - For luminance planes:
> > +     *
> > +     *       (219 * E + 16) * 2^(n-8)
> > +     *
> > +     *   F.ex. the range of 16-235 for 8 bits
> > +     *
> > +     * - For chrominance planes:
> > +     *
> > +     *       (224 * E + 128) * 2^(n-8)
> > +     *
> > +     *   F.ex. the range of 16-240 for 8 bits
> > +     */
> >
>
> I might have a minor suggestion: instead of giving the example for 8 bit
> only, why not provide the generic formula for any bitdepth?

I think I already have the formulas there ;) .

If you generate the doxygen, it even puts the formula into a nice code block.

> If you think it's too convoluted or the wrong place for this information,
> I'd still recommend adding an example for 10 bit too, since a lot of new
> content is produced in 10 bit, and could be a useful reference.
>

I think personally a single example is enough since the formulas and
definition of E and n are already there. But if anyone else feels
heavily for adding the 10 bit example values, I can add them.

Jan