[Bf-vfx] Improving the 2D stabilization tool :: Next Steps

[Ichthyostega]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


Seems both Sergey and I agree that it's best to wrap up the current improvements
first. Also I agree that camera distortion is not a topic for the 2D tool

Thus IMHO the next steps would be

* code review and any offspring from there

* solving the remaining problems: animation / correct compositor output

Then, I'd like to address the following things

* finding a way to use automation values correct in initialisation and
  calculation of the autoscale-factor

* making the autoscale function more useful: turn autoscale in some kind of
  automation driver
  - calculate autoscale only for a range of frames (not full clip)
  - use autoscale result to calculate a starting point for "target position


Beyond these near term goals, I consider the following topics important:

- - what are the components of "autoscale data"?
  position + rotation + scale + pivot point ?

- - how to handle and possibly calculate the pivot point better

- - how does the pivot point integrate with compositor?



Am 05.06.2014 08:10, schrieb Sergey Sharybin:
> - Not sure why animated weight is not applied in compositor. It should work
>  just fine, seems i'll need to have a closer look into the code. And yes,
> it's not editable in animation editors. This is something to implement
> still.

I have investigated that a bit more. I added a movie clip editor to the
Compositor screen. In the MCE, I can see the animated target position
and track weight properties.

But strange enough, neither the MCE view, nor the backdrop in compositor
shows the effect of those target position parameters. When I switch to the
tracking screen, on the same frame the effect of these parameters is visible.

This is especially puzzling, since I work in these parameters right into the
data produced by the stabilizer. Thus it is unlikely that other parts of
Blender would get the raw stabilization data without target position worked
in. Thus I suspect that actually the transformation as viewed in Compositor
is somehow not wired up in the right way. Since I had to tweak and reorder
the matrix to get correct results in the MCE. And as far as I can see,
the compositor doesn't use that matrix, but builds his own transformation
based on the stabilization data feed.


Cheers,
Ichthyo



-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (GNU/Linux)
Comment: Using GnuPG with Icedove - http://www.enigmail.net/

iEYEARECAAYFAlORNRQACgkQZbZrB6HelLLgiACgiCe7IfndJPS6OPZ7T5+J+Elb
jpkAnAgnMxVrGRgk0zkx1h+5AUhUws3H
=Fz3F
-----END PGP SIGNATURE-----