- Switch from gaussian-bokeh blur to mean-value which is much faster
and from matte's quality behaves not worse than gaussian-bokeh.
- Added Edge matte output so areas where black/white clipping happens
could be easily controlled.
Currently not sure if it's possible than operation returns two
buffers, so now implemented as separate clipping operation call
with specific options.
- Changed logic of detecting edges - check if most pixels within
kernel are different fits some tolerance interval.
- Added two extra options which controls how edges are detecting:
* Edge Kernel Radius defines radius of kernel within pixel's value
is checking.
* Edge Kernel Tolerance defines maximal difference between value
of current pixel and it's neigbours which are trated as belng to
the same plane )background of foreground)
Unfortunately, value for this two sliders for existing files shall
be tweaked manually. Default values are 3 for radius and 0.1 for
tolerance.
Problem was in fact that non-linear-contiguous axis assignement was broken (i.e. location.xy would work as expected, but location.xz would only affect .x part)... Now all possibilities should work fine.
Did not try to fix the problem specific to obj.dimension (when assigning multiple times to this array, only the last one is taken into account - in fact, a simple print() shows that assigning to dimension is not taken into account immediately), not sure whether this is normal behavior, or if we need a specific "update" of some kind for this prop?
- new compositor could use uninitialized var
- profile conversion could use uninitialized var
- set better warnings for clang+cmake.
- remove picky warnings from sphinx doc gen shell script.
Looks like drawing grease pencil before masks was affecting projection matrix
somehow which made masks invisible
Anyway, drawing GP actually shall happen after masks to match how it works
in other areas.
Helps keeping features tracked when there's large scale happens
without need to manually re-adjust search area.
Currently using factor of pattern's boundbox scale, but probably
could be done in more accurate way?
artifacts in the boundaries of tiles.
Also added test blur operation which could potentially give better results
than gaussian blur for keying. Not user yet, to enable uncomment like
with USE_GAUSSIAN_BLUR in file COM_KeyingNode.cpp.
Added selection flush after loop cut, changed select_linked_pick's
flush to use the em selection flag.
Fixes bug [#31715] Cases where verts and edges are selected but not
relevant faces
Add check for merging vertices into vertices that are themselves
marked for merge, was already done for array eleements but not end
caps.
Fixes bug [#31695] Array Modifier: End Cap fails if all vertices are merged
Also corrected some reversed assert arguments.
Bug [#31785] Applying a transform to an object with multires weird result
Was reading interleaved coord/mask data incorrectly since paint mask
merge. Fixed by using two separate CCGKeys.
Some additional code cleanup: deduplicate multires tangent matrix
calculation.
only visible when doing large size blurs.
also solved the catcom/mitch filter that didn't work at low/medium quality
PS never use BokehBlur Gausian filter as it is 99%+ identicat as non bokeh <= top for sergey- :)
AFAIK, it is impossible to determine exactly which axes may have negative
scaling values from a 4x4 matrix (which is the underlying cause of this bug).
However, we can figure out if there is some negative scaling going on in that
matrix (i.e. one of the axes has negative scale). So, the fix here is to
negatively scale everything if we detect this happening.
WARNING: do not rely on being able to accurately detecting positive/negative
values for more than a single axis per bone controller. Weird results may occur.
You have been warned.
Slightly hacky fix to get updates working for Action Editor header when there
are changes of the active action (via undo and/or deleting all keyframes at
once). Since the action referenced by the editor only gets updated when anim
channel filtering (e.g. as a result of the channel syncing operation) is
invoked. Added comments noting where these updates actually occur
space works again
This commit restores the support for using Grease Pencil in the Sequence Editor
image preview region, making it possible to scribble on footage for review
purposes again. Due to internal changes in how the Sequencer handles the image
drawing for this stuff (i.e. it is now fully based on View2D instead of trying
to implement its own little crazy offset+zoom stuff), a lot of the old code for
handling those offsets is no longer needed. Instead, one of the "standard" cases
is now used, and works quite well.
Bugfixes:
* View-space Grease Pencil drawing was done in wrong place (before view2d
restore)
* Grease Pencil entry in RNA had wrong/missing type
Credits:
* DingTo - initial patch/attempt at restoring support
* Aligorith - solved the "offset problems"
