[FFmpeg-user] Generating a displacement map

[Evert Vorster]

Hi there, expert ffmpeg users.

Back story:
Samsung Gear360 camera produces 4K HEVC encoded video clips, but puts the
raw input from two sensors on to one frame.
The sensors are behind fisheye 190 degree glass, so the images in each
frame shows up as two circles. (shperes?)
Using ffmpeg, hugin, and some scripting, each image is un-distorted
according to it's lens characteristics, and stitched together to make an
equirectangular frame.
The template does not change from frame to frame, but it does take a long
time to chew through thousands of frames, and the quality loss is
horrible.

tl,dr;
Each pixel in the input frame end up somewhere on the output frame in a
predictable way.


On to the question:
Rather than figure out the location of each pixel every time, it would be
much more efficient to just use a displacement map and move the pixels.

Actually generating such a map might be a bit of a trick.
Is there any ideas from the community on how to generate and use such a map?

Kind regards,
-- 
Evert Vorster
Isometrix Acquistion Superchief