[FFmpeg-devel] [PATCH] avfilter: add panorama filter

[Paul B Mahol]

On 3/12/18, Hazem Ashmawy <hazem.s.ashmawy at gmail.com> wrote:
> On 3/12/18, Paul B Mahol <onemda at gmail.com> wrote:
>> On 3/12/18, Hazem Ashmawy <hazem.s.ashmawy at gmail.com> wrote:
>>> So, I spend few hours trying to incorporate the partial change.
>>>
>>> I'm not sure that the video sent by <jamrial> earlier follows the
>>> faces order here
>>> https://github.com/google/spatial-media/blob/master/docs/spherical-video-v2-rfc.md#semantics-3
>>>
>>> Because when I made this change, I got a reasonable equirectangular
>>> ouput:
>>> https://imgur.com/a/BEN00
>>>
>>> comparing to the changes sent earlier.
>>> https://imgur.com/a/vaIvW
>>>
>>> diff --git a/libavfilter/vf_panorama.c b/libavfilter/vf_panorama.c
>>> index de08ef4..5383d57 100644
>>> --- a/libavfilter/vf_panorama.c
>>> +++ b/libavfilter/vf_panorama.c
>>> @@ -34,12 +34,12 @@ enum Projections {
>>>  };
>>>
>>>  enum Faces {
>>> +    DOWN,
>>>      LEFT,
>>> +    TOP,
>>>      FRONT,
>>>      RIGHT,
>>> -    TOP,
>>>      BACK,
>>> -    DOWN,
>>>  };
>>>
>>>  struct XYRemap {
>>> @@ -403,7 +403,7 @@ static int config_output(AVFilterLink *outlink)
>>>
>>>                      switch (face) {
>>>                      case LEFT:
>>> -                        locate(z, x, y, M_PI,   rw, rh, &ox, &oy);
>>> +                        locate(z, x, y, -M_PI_2,   rw, rh, &ox, &oy);
>>>                          break;
>>>                      case FRONT:
>>>                          locate(x, z, y, 0.,     rw, rh, &ox, &oy);
>>> @@ -418,7 +418,7 @@ static int config_output(AVFilterLink *outlink)
>>>                          locate(x, y, z,-M_PI_2, rw, rh, &ox, &oy);
>>>                          break;
>>>                      case DOWN:
>>> -                        locate(y, x, z,-M_PI_2, rw, rh, &ox, &oy);
>>> +                        locate(y, x, z, M_PI_2, rw, rh, &ox, &oy);
>>>                          break;
>>>                      }
>>>
>>> What I'm saying is: when I changed face order to this (which video
>>> sample seems to follow), I got a reasonable output.
>>> down | left | top
>>> front | right | back
>>>
>>> Comparing to Google's:
>>> right | left | top
>>> down | front | back
>>
>> The video linked here does not have name of faces, so you can not know
>> what is front and what is back.
>>
>> Just add support for typical cubemap layout currently used.
> You mean In the video linked?
> Or the Google's specification and ignore the linked video?
> Or current layout order originally used in the patch?

Just use what currently cubemap layout is used by youtube.