Re: [FFmpeg-devel] [PATCH] libsvtav1: Add workaround for gop_size == 1

From: "Ronald S. Bultje" <rsbultje@gmail.com>
To: FFmpeg development discussions and patches <ffmpeg-devel@ffmpeg.org>
Subject: Re: [FFmpeg-devel] [PATCH] libsvtav1: Add workaround for gop_size == 1
Date: Tue, 27 Jun 2023 15:07:00 -0400
Message-ID: <CAEEMt2nzfF92dDfpf6v9TRiPe8Y2uSguKdjoCY9YkqsuRBSYbw@mail.gmail.com> (raw)
In-Reply-To: <CAOJaEPKEdn6AX5owBc+WzB=2kKo1Sw6hf3sURUV0MeY0GLC_5w@mail.gmail.com>

Hi Vignesh,

On Tue, Jun 27, 2023 at 1:55 PM Vignesh Venkat <
vigneshv-at-google.com@ffmpeg.org> wrote:

> > >>> In some versions of libsvtav1, setting intra_period_length to 0
>
[..]

> > >>> SVT-AV1 Bug: https://gitlab.com/AOMediaCodec/SVT-AV1/-/issues/2076
> > >>>
> > >>> Example command: ffmpeg -f lavfi -i
> testsrc=duration=1:size=64x64:rate=30 -c:v libsvtav1 -g 1 -y test.webm
> > >>>
> > >>> Before: Only first frame is keyframe.
> > >>> After: All frames are keyframes.
>
[..]

> From what i tested, it seems to be there on the tip of tree SVT-AV1 as
> well. So we can't guard it with macros until the issue is fixed.
>

Isn't this the expected behaviour, btw?

See https://gitlab.com/AOMediaCodec/SVT-AV1/-/blob/master/Docs/Parameters.md

"Keyint --keyint [-2-(2^31)-1] -2 GOP size (frames), use s suffix for
seconds (SvtAv1EncApp only) [-2: ~5 seconds, -1: "infinite" only for CRF,
0: == -1] "

note the final parts: 0=-1, and -1 means "infinite" keyint (i.e. never
insert a new KF). Since FFmpeg's wrapper sets keyint to "gopsize value
minus 1", "-g 1" seems to become equivalent to "--keyint 0" on the SVT-AV1
commandline, unless there's some discrepancy between API and CLI interface.

(I admit that if I set it to 1, I don't actually get keyframes, but rather
intraonly frames for all except the first frame. --keyint 2 correctly uses
key - inter pairs. But your "before" statement above appeared to suggest
that the first keyframe is followed by inter frames, not intraonly frames.
If you meant "intraonly after key" instead of "inter after key", you should
probably note that more explicitly. This may also help upstream reproduce
in #2076 so they can fix it going forward.)

Ronald
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".