The issue was caused by uncertainty of current camera when
there're no markers to the left of current frame.
Now in this case camera from the top-left marker will be used.
My previous fix for uninitialized texture coordinates was not working well,
and in fact there was a bigger issue with GLSL drawing and missing attributes
with immediate draw mode. Now it will explicitly pass zero rather than having
it use whatever value was set last.
It was rather confusing from the user usage point
of view and didn't get so much improvement after
new bundle adjuster was added.
In the future we might want to switch resection
to PPnP algorithm, which could also might be a
nice alternative to fallback option.
The way node groups check for localized trees in the ntreeFreeTree_ex function does not work. When the main library is freed on exit it also frees genuine node groups trees (which is correct), but then
node groups referencing these trees will not find them in the library and interpret that as a localized group, attempting to free them a second time ... Nicer solution is to just use a special flag on
localized node trees so we can clearly distinguish them from genuine trees in main.
This option is needed when vertex groups are used for both armature
deformation and the other purpose such as influence of mesh modifier.
Thanks to Campbell for code review!
Was a silly error introduced in algo during optimization, vertices having sharp and smooth edges would only get the first poly's normal for each "smooth loop group".
Also done a (minor) optimization, and some comment fixes.
We now support the combined layer of Photoshop files (stored as layer 0
in the file). This way users can keep their files as multilayer PSD and
Blender always handle them as flat images.
For perfect alpha this requires an OpenImageIO update:
https://github.com/OpenImageIO/oiio/commit/342cc2633ff590a3bb278481c61ae798c7148361
Photoshop sample files:
https://github.com/OpenImageIO/oiio-images
Brecht has some pending fixes to push for OIIO as well, so we may as
well wait to update our libraries.
What works:
===========
* 8bit images (with or without alpha)
* 16bits images (alpha discarded)
* Photoshop files saved with 'Maximum Compatibility'
* Cycles, Blender internal, BGE (and player)
Known limitations
(due to OIIO dependency):
=========================
* Images with less than 4 channels show a wrong thumbnail (bug may be in OIIO)
* Packed images are not supported
* We do not write PSD files.
Note: old Blenders have support for PSD via Quicktime library. But due
to license issues this was discontinued.
Many thanks for Brecht van Lommel for reviewing the patch, suggesting
multiple improvements and to help solving the alpha issue.
The SAMPLELINE flag in histogram was set during the BKE_histogram_update_sample_line function. That function in turn is called during every scope update in area draw function, meaning that during render
it constantly gets set. OTOH the operator tries to disable the flag on invoke, which "cancels" the sample line by default. So during render the operator un-setting of the flag has no effect, because the
render job immediatly triggers a redraw, which updates scopes and sets it again.
Moved the flag out of the actual sample line update function into the operator execute. Now only the operator enables/disables overall sample line drawing, while the rest of the update works as before.
* The mask was not subdivided properly on splitting edges, which gave interesting
but definitely wrong fractal-like borders around masks.
* Edge splitting was only done where the mask was < 50%, with the reasoning that
you can't do a 50% topology update. But this gives an ugly border in the mesh.
The mask should already make the brush move the vertices only 50%, which means
that topology updates will also happen less frequent, that should be enough.
The "active ID node" concept has become slightly more complex with pynodes. To find the active material or other ID links in a node tree recursively requires a hash key based on the "parent" tree of the
current node group. To avoid returning NULL in case this key is not yet initialized (i.e. ID node has not been activated yet), just accept 0 key as well for the base node tree.
remove a vertex from the pbvh if the only face it was part of was
removed from the pbvh. This should work but no time to investigate now
properly, and better have a solid release. The problem could be seen
when sculpting to open meshes. Some vertices and faces on the boundaries
would become stuck and unresponsive.
data blocks are freed using the standard ntreeFreeTree function, so freeing localized node groups has to be done there. This means an ugly loop over G.main to detect localized groups, but should not be a
big problem in practice.
into main or is an independent data block. This can now be controlled by passing an explicit Main pointer, so we can ensure localized node groups don't pollute main.
This problem was introduced with pynodes merge in r55373. It's caused by missing localization of node groups in shaders in cases where GLSL + node previews causes threading conflicts. I'm not quite sure why
we didn't do this before, but now all node groups also get localized recursively.
Meta sound update (seq_update_sound_bounds_recursive_rec) was not taking into account hard trim (anim_startofs) when setting sound's start, while default sound strip update (sound_move_scene_sound_defaults) did... This could use some refactor, though, with a single func used in both cases, to avoid such issue.
Also added soft trim to sound panel, only hard one was available.
Jittering was caused by homography not being estimated
accurate enough.
Before this, only algebraic estimation was used, which
is indeed not so much great, Now use algebraic estimation
followed with refinement step using Ceres minimizer.
The code was already there since keyframe selection patch,
made such estimation a generic function in multiview/ and
changed API for estimation in order to pass all additional
options via an options structure (the same way as it's
done fr Ceres).
This includes changes to both homography and fundamental
estimation.
TODO:
- Need to document Ceres functors better.
- Need to support homogeneous coordinates (currently
only euclidean coords are supported).
Sergey Sharybin
Brecht Van Lommel
Campbell Barton
Lukas Toenne
Shinsuke Irie
Bastien Montagne
Antonis Ryakiotakis
Tamito Kajiyama