[FFmpeg-devel] Vignette filter limitations...

Evert Vorster evorster at gmail.com
Sun Aug 27 10:33:30 EEST 2017


Hi there.

I have asked this question on the ffmpeg-users list, but it's been oddly
quiet on this front.

I am trying to create a ffmpeg command line that re-maps and stitches
together footage from the Samsung Gear 360 camera. The basic methodoly can
be applied to any 360 view camera. The reason I am doing this is that the
software that is available to do this is closed source and extremely
expensive. For a hobbiest, this is a bad situation.

First thing I do, is to load a specially crafted series of frames in Hugin,
and perfectly map the lenses.
Then I make the remap files that the ffmpeg remap filter uses with nona -c.
I hand craft a alpha map to do the blending between the lenses, and the
results are really good.

Unfortunately I have a real problem with vignetting.
The vignette filter in ffmpeg seems to have some room for improvement.
In Ffmpeg there is only one variable, and the x,y center. This allows for
only one type of vignetting correction with the "backward" option set.

In Hugin (and the panotools) the lens is described with 3 variables, and an
x,y center. This enables the vignetting strength to be described as a
custom curve.

If you are curious about the project, this is the github page, with
examples.
https://github.com/evertvorster/dualfisheye2equirectangular_ffmpeg_remap
There are example .pto files, that when loaded in Hugin show the lens
chataristics of the Samsung Gear 360 lenses, and some sample footage to
test on.

How difficult would it be to port the panotools' vignetting correcting
filter into ffmpeg?
-- 
Evert Vorster
Isometrix Acquistion Superchief


More information about the ffmpeg-devel mailing list