[vtk-developers] Class design for spline visualizations

[Lin M]

Hi Dr. Thompson,

I have written a helper function to generate control points for arbitrary
ellipse and modified the PatchInterpolation.py test to image test which
contains line, circle and ellipse.

A question about the push the data. I followed the instruction in
https://gitlab.kitware.com/vtk/vtk/blob/master/Documentation/dev/git/data.md,
but I didn't see the message when doing the commit step.

Some/Module/Testing/Data/Baseline/MyTest.png.md5: Added content to Git
at refs/data/MD5/...
Some/Module/Testing/Data/Baseline/MyTest.png.md5: Added content to
local store at .ExternalData/MD5/...
Content link Some/Module/Testing/Data/Baseline/MyTest.png.md5 ->
.ExternalData/MD5/...

The image test runs correctly in my local machine.

Best,
Lin

On Sat, May 23, 2015 at 12:16 AM, David Thompson <david.thompson at kitware.com
> wrote:

> Hi Lin,
>
> I believe VTK expects each test to generate a single baseline image whose
> filename matches the Python script name (e.g.,
> TestSynchronizedTemplates3D.py
> <https://gitlab.kitware.com/vtk/vtk/blob/master/Filters/Core/Testing/Python/TestSynchronizedTemplates3D.py> must
> have its baseline image named TestSynchronizedTemplates3D.
> <https://gitlab.kitware.com/vtk/vtk/blob/master/Filters/Core/Testing/Python/TestSynchronizedTemplates3D.py>png
> in the Data/Baselines directory).
>
>     David
>
> On May 22, 2015, at 23:15, Lin M <majcjc at gmail.com> wrote:
>
> Hi Dr. Thompson
>
> I changed the python test to baseline image and followed the instructions
> in the VTK Data page, but running CMake doesn't seem to make the .md5
> files. Do you have any idea about that?
>
> I outputed two images named bezier-circle.png and bezier-line.png and put
> them in source-tree/Filters/Bezier/Testing/Data/Baseline.
>
> Best,
> Lin
>
>
> On Tue, May 19, 2015 at 1:52 PM, David Thompson <
> david.thompson at kitware.com> wrote:
>
>> Hi Lin,
>>
>> A few additional things that might help:
>>
>> 1. When you are done reviewing the merge request I assigned to you,
>> please click "Accept merge request" or add a "+1" comment so that it will
>> become part of the splines/filter-bezier branch.
>>
>> 2. A good pattern to follow for the image-based test is
>> Filters/Sources/Testing/Python/TestPlatonicSolids.py . You can see how it
>> obtains the path to img_file and calls vtk.test.Testing.compareImage to
>> perform a rendering to compare. The baseline image that the rendering is
>> compared against is not stored in the VTK git repo. Instead, MD5 checksums
>> of images are stored in the git repo and the images are uploaded
>> separately. See
>>
>>
>> https://gitlab.kitware.com/vtk/vtk/blob/master/Documentation/dev/git/data.md
>>
>> for a description of how to generate a test image, add its MD5 sum, and
>> upload it with your merge request.
>>
>> 3. If you are going to generate helper methods to create conic sections,
>> then be aware that you'll have to return multiple sets of control points
>> because ellipses cannot be represented with a single patch (it is best to
>> return one patch per quadrant) because of the way they are parameterized.
>>
>>         David
>>
>> > On May 18, 2015, at 10:45 PM, Lin M <majcjc at gmail.com> wrote:
>> >
>> > Hi Dr. Thompson
>> >
>> > Thanks! I'll look into that and implement the test asap.
>> >
>> > Best,
>> > Lin
>> >
>> > On Mon, May 18, 2015 at 10:36 PM, David Thompson <
>> david.thompson at kitware.com> wrote:
>> > Hi Lin,
>> >
>> > Sorry for the slow reply.
>> >
>> > > Is there any dataset for precomputed Bezier interpolation? I'm
>> currently comparing the result with my manual calculating. In this way I
>> can not test many examples and it's quite low efficient.
>> >
>> > I've pushed a merge request to the project that adds a couple simple
>> test cases. I'll add some more over the next few days. You can run the test
>> by building VTK with the VTK_WRAP_PYTHON option turned ON and then running:
>> >
>> >     make
>> >     ./bin/vtkpython
>> /path/to/src/Filters/Bezier/Testing/Python/PatchInterpolation.py
>> >
>> > If you uncomment the lines below the "For debugging, ..." message in
>> PatchInterpolation.py, running the test will create some files (named
>> bezier-circle.vtk and bezier-line.vtk) that you can load into ParaView.
>> >
>> > 1. As I mentioned earlier, it would be great to do image-based testing.
>> A good exercise would be to modify the Python test into an image test
>> version.
>> >
>> > 2. Another good exercise would be to generate control points for other
>> conic sections such as hyperbolas, parabolas, and the more generate case of
>> ellipses. I recommend "The NURBS Book" by Les Piegl; the circle quadrant
>> example in the Python test is from pp. 26--32 of the book. It would be
>> really nice to have a utility that would create control points given a few
>> shape parameters. For example:
>> >
>> >     ctrlPts, prange = ellipse(major_radius=2, minor_radius=1)
>> >     ctrlPts, prange = hyperbola(semi_major_axis=2/3, eccentricity=1)
>> >
>> > Once we have the 1-, 2-, and 3-d Bezier interpolation tested, we can
>> work on converting B-splines into Bézier patches and testing that.
>> >
>> >         David
>> >
>> > >
>> > > On Mon, Apr 27, 2015 at 1:46 PM, David Thompson <
>> david.thompson at kitware.com> wrote:
>> > > Hi Lin,
>> > >
>> > > > ... I changed the method to compute binomial coefficient from
>> vtkMath::Binomial to vtkMath::Factorial using our memoization improvement
>> and tried to merge from Lin.Ma/filter-bezier to Spline/filter-bezier.
>> > >
>> > > Great! I can see that you were able to merge to the branch.
>> > >
>> > > > What do you think is the proper thing to do next?
>> > >
>> > > Right now, it looks like TestPatchInterpolation is modifying the
>> array holding control point coordinates. I think it would be a good idea to
>> write the interpolated points to a separate array and then make the test
>> verify that Bernstein polynomial is being computed properly.
>> > >
>> > > There are 2 ways to verify things: value tests and image tests.
>> > >
>> > > 1. Value testing. Interpolate a few different r values on a few
>> different sets of control points and compare the resulting point
>> coordinates to values you know are good. The comparison must allow for
>> small differences due to the differences in floating-point arithmetic
>> implementations on different platforms.
>> > >
>> > > 2. Image testing. Plot the interpolated points and compare the
>> rendering of the interpolated points to one you have verified is correct.
>> While not as accurate as value testing, it can often be easier to debug a
>> failing test with a rendered image. VTK provides utilities for comparing
>> images (also allowing for small differences due to floating-point math,
>> OpenGL implementations, and so on). The image baselines that get compared
>> to your test's rendering are stored outside of the git repository to keep
>> the repository from getting bloated. Instead, the MD5 sum of the baseline
>> images are stored in the git repository and the images are stored
>> separately on Kitware's servers, which can be queried to find a file with a
>> given MD5 sum.
>> > >
>> > > Please take a look at this documentation on adding test data and
>> baseline images to VTK:
>> > >
>> > >
>> https://gitlab.kitware.com/vtk/vtk/blob/master/Documentation/dev/git/data.md
>> > >
>> > > and see if you can add an image-based test of the patch interpolation.
>> > >
>> > >         David
>> > >
>> >
>> >
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/vtk-developers/attachments/20150530/68eb4a4b/attachment.html>