[FFmpeg-devel] [PATCH] libavcodec/libfdk-aacenc: Scale VBR mode with FF_QP2LAMBDA

[Jonathan Gee]

I had noticed fdk is specifically looking for the qscale flag, and
otherwise does not do anything with global_quality.  I suppose the change
is risky for anyone who is setting both global quality and qscale, but with
the current code, it seems incorrect to have a conditional based on the
scaled option and not account for the scaling.

if (avctx->flags & AV_CODEC_FLAG_QSCALE || s->vbr) {
    int mode = s->vbr ? s->vbr : avctx->global_quality;

On Wed, Feb 22, 2023 at 9:16 AM Timo Rothenpieler <timo at rothenpieler.org>
wrote:

> On 22.02.2023 17:46, Gyan Doshi wrote:
> >
> >
> > On 2023-02-22 10:12 pm, Timo Rothenpieler wrote:
> >> On 22.02.2023 17:33, Gyan Doshi wrote:
> >>> From
> >>>
> >>> fftools\ffmpeg_mux_init.c
> >>> 619:        ost->enc_ctx->global_quality = FF_QP2LAMBDA * qscale
> >>>
> >>> Regards,
> >>> Gyan
> >>
> >> But that's only if you set the old qscale CLI options.
> >> If you set the global_quality option directly, there is no factor
> >> applied.
> >>
> >> By dividing by FF_QP2LAMBDA you break setting this value via
> >> global_quality, and instead add support for setting it via the old and
> >> pretty much abandoned qscale interface.
> >
> > FWIW, that's what LAME does.
> >
> > libavcodec\libmp3lame.c
> > 119:        lame_set_VBR_quality(s->gfp, avctx->global_quality /
> > (float)FF_QP2LAMBDA);
> >
> > global_quality semantics seem overloaded. Maybe we should just redirect
> > user to priv vbr  and error out in fdk init.
> >
> > Regards,
> > Gyan
>
> It's pretty much a matter of "what did this code always do in the past".
> It then got to stick to it, cause otherwise we break downstream API and
> CLI consumers.
>
> The mp3lame code is probably old enough that qscale was still the
> default at the time.
> Nowadays it's global_quality.
>
> Both of those options mapping to the same field in avctx, one with a
> magic factor applies, is definitely and oddity.
> _______________________________________________
> 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".
>