[FFmpeg-devel] Revert a patch for v360 filter

[Paul B Mahol]

On Fri, Oct 8, 2021 at 2:11 PM Michael Koch <astroelectronic at t-online.de>
wrote:

> I'd like to suggest to revert this patch for the v360 filter:
> http://ffmpeg.org/pipermail/ffmpeg-cvslog/2020-October/124887.html
>
> Why should it be reverted?
> -- Because after this patch, the yaw, pitch and roll angles have a
> different meaning. If they are set in the command line, they are
> interpreted as absolute angles. However if they are sent via sendcmd,
> they are interpreted as relative angles.
> -- As far as I know, no other filter has such behaviour where an option
> gets a different meaning if it's set via sendcmd.
> -- It's impossible to test a rotation in the command line, and later set
> the same rotation via sendcmd. The result would be totally different.
> -- It breaks previously correct behavior. See for example chapters 2.111
> and 2.115 in my book:
> http://www.astro-electronic.de/FFmpeg_Book.pdf
> -- It's undocumented.
> -- I don't see any workaround to simulate an absolute rotation with
> relative rotations. I've tried it but it doesn't work. It might work for
> one rotation axis, but if two or three axes are involved simultaneously,
> it gets extremely complicated.
>
> Why was this patch made?
> -- Because there was a problem with gimbal lock in some rare cases. I
> never saw this problem.
>

Invalid reasoning. Absolute rotations give those issues.
Using linear interpolation for absolute or relative rotation is flawed
approach
and give various broken results for non-trivial rotations.
Filter is meant to be used interactively and not via sendcmd.



> What's the correct way to solve this problem?
> -- Most important is to revert this patch. That means there will be
> gimbal lock in some rare cases. The user has to take care of it, and
> avoid these rare cases.
> -- New options could be added for relative rotations: rel_yaw,
> rel_pitch, rel_roll. Then the user could decide if he prefers absolute
> or relative rotations. However from my point of view, I don't need
> relative rotations.
>
> See also ticket 9447.
>
> Michael
>
> _______________________________________________
> 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".
>