Blender Git Commits

Merging r58947 through r58955 from trunk into soc-2011-tomato

Merging r58870 through r58946 from trunk into soc-2011-tomato

Code cleanup: made function name consistemt

Remove downsample operation

It didn't work too much nice, and finishing it
is not so high priority for now.

Files would be in svn history anyway, so if
someone would like to look into the code, he's
welcome to do this.

Remove plane track mask warping operation

After some thoughts, it's not needed actually.
As soon as mask is parented to a plane, Mask
input node would give you warped mask already.

Made corner sliding more convenient

Now it's possible to slide corner roughly, then
press Shift to switch to accurate positioning.

Before this change corner used to jump when one
presses shift, which wasn't too much convenient.

Fix wrong usage of mul_v3_m3v3.

You couldn't use the same vector as input and output.

Use a proper depending are of interest for image warping

Instead of requesting the whole frame to be evaluated
before plane warping could be invoked, warp tile into
original image space and use it's bounding box as area
of interest.

There're some tricks with margins going on still, but
without them some tiles are missing.

Quick tests seems to be working fine, we could solve
possible remaining issue later.

Replace crappy custom sampler with EWA one

Our sampler could be useful for more general usage,
like simpler downsampling in scale node, but it needs
to be improved before it's really useful.

It was giving lots of jittering artifacts which isn't
good for motion tracking. So now image wrap operation
used EWA filtering.

For now, it uses copy-pasted dx/dy calculation which
came from MapUV operation. Ideally we need to get
rid of duplicated code, but it's not so trivial for
now because of UV coordinates are calculating in
different ways. Not a big deal to have duplicated
code for a while.

Initial code layout for plane track deform node

Idea of this is:
- User selects which plane track to use (for this he
need to select movie clip datablock, object and track
names).
- Node gets an image and mask inputs (both are optional).
- Node outputs:
* Input image warped into the plane.
* Input mask warped by the plane.
* Plane, rasterized to a mask.

Warping image is done by computing reverse bilinear
coordinates, and getting pixel from corresponding
position.

This requires some tricks with downsampling to make warped
image looks smooth.

Currently compositor doesn't support downsampling, so we
needed to implement our own operation for this.

Currently idea is dead simple: value of output pixel equals
to an average of neighborhood of corresponding pixel in input
image. Size of neighborhood is defined by ratio between input
and output resolutions.

This operation doesn't give perfect results and works only
for downsampling. But it's totally internal operation not
exposed to the interface, so it's easy to replace it with
smarter bi-directional sampler with nicer filtering.

Limitations:
- Node currently only warps image and outputs mask created
out of plane, warping input mask is not implemented yet.
- Image warping doesn't report proper depending area of
interest yet, meaning interactivity might be not so much
great,
- There's no anti-aliasing applied on the edges of warped
image and plane mask, so they look really sharp at this
moment.

Merging r58795 through r58869 from trunk into soc-2011-tomato

Compute homographies when creating new plane tracker

When new plane tracker is creating, blender will go
over tracked sequence of point tracks, compute homography
between neighbor frames and deform plane rectangle using
this homography.

This makes rectangle plane follow tracked plane quite
nicely. But as an additional homographies will be
updated after corners were slided with a mouse.

Additional changes:

- Display keyframed/tracked information in the cache line
for plane tracks (the same way as it's done for point tracks).
- Fix couple of buys in homography C-API.

Add Procrustes PNP ("PPnP") resection algorithm to libmv

This adds a new Euclidean resection method, used to create the
initial reconstruction in the motion tracker, to libmv. The method
is based on the Procrustes PNP algorithm (aka "PPnP"). Currently
the algorithm is not connected with the motion tracker, but it
will be eventually since it supports initialization.

Having an initial guess when doing resection is important for
ambiguous cases where potentially the user could offer extra
guidance to the solver, in the form of "this point is in front of
that point".

Initial code layout for real plane tracker

This commit includes:

- DNA structures layout. It's not completely finished
yet, and at some point files could become broken,
but we'll try to avoid this as much as we can.

- Basic RNA code layout for new data structures.
Not completely finished in terms it's not possible
to define plane tracks from Python just yet.

- Operator to define the plane.

- Deletion and selection operators are aware of planes.

- Plane marker's corners could be slided with mouse.

Still lots of thing to be done, but we need to commit
at some point so we could share the code work.

Fix compilation error after recent merge.

Merging r58363 through r58794 from trunk into soc-2011-tomato

Remove files which were actually removed form trunk already.

Remove workarounds for mask depsraph issue,

They're not needed after recent trunk merge.

Merging r57346 through r57354 from trunk into soc-2011-tomato

Merging r57134 through r57345 from trunk into soc-2011-tomato