[FFmpeg-user] Problem in v360?

Michael Koch astroelectronic at t-online.de
Sun Oct 25 10:05:22 EET 2020


Am 25.10.2020 um 03:02 schrieb Edward Park:
> Hi,
>
>> After a deeper look into the quaternion multiplication, it seems to be correct. The terms are only written in a different order in the Wikipedia article.
>>
>> But it's a fact that the v360 rotations are broken, as can be shown by this simple test:
>> You can use any equirectangular input image.
>> First make a combined rotation yaw=90 and pitch=45 in the default rotation order (ypr),
>> then rotate back pitch = -45,
>> then rotate back yaw = -90.
>> The output image should be the same as the input image. But it isn't.
>
> It seems to me it assumes a 45° vertical fov and 90° horizontal fov on both input and output, does that affect the behavior at all? (Or am I not understanding right)

If the input and output images are specified as "equirectangular" 
(V360=e:e:...) then the field of view is always 360° x 180°. There are 
options for for input and output field of view, but these aren't used in 
the case of equirectangular images. Something seems to be wrong with the 
quaternions that were added October 7th.
This is difficult stuff. A wild guess that might be worth testing:
Swap the order of quaternion multiplication, 2nd and 3rd argument of 
multiply_quaternion()

Michael



More information about the ffmpeg-user mailing list