I can tell you that because the action is ‘fast’ and
the frame differences are huge, it’d be necessary
to guide Twixtor’s motion detection using some
manual tracking (in Twixtor Pro) at least to separate
certain ‘areas’ from each other. Plus the clip has
been captured with a fairly long exposure time
(hence motion blur). So there’s no one magical
setting that would track everything fully, and
the manual tracking work would be the thing to do.
You could easily spend several hours to set up polys
for something like this. If you’re in a hurry, you
may want to concentrate on the hands and fingers
making them the ‘foreground’ objects, since
there are less details and interest in his gi.
His face would be easy to track, except where his
hands go ‘in the foreground’ and when he starts to
rotate towards the camera.
Perhaps another approach might be to extend where there’s
less frame differences and quicken where there are –
to exagerrate both the ‘fast’ parts and ‘slow’ parts –
using frame-number speed control. For such a speed
ramp, the motion blur compensation would be quite
effective since it will exaggerate blur where
action is ‘fast’ and subtract it where the action is
‘slowed down’.
HTH
-Shin