Import the algorithms present in imagej-ops #1

tinevez · 2023-07-03T13:05:34Z

This break the separation between the data structure and the algorithms. But there are reasons why we want to do this:

facilitate development of data structures without breaking ops
offer to consumers one lib that has algos and data structures
offer consumers not to have to pull all imagej-ops if they just want to play with meshes
if there will be an imagej-ops2, then it can have default implementations that link to the algos in this lib, instead of copying the code from imagej-ops1.

ctrueden

@tinevez Thanks for all your hard work on migrating all this goodness into one place! 🥇

I made a variety of suggestions. But the big ones I think are:

The PR won't merge as is, needs a rebase first to fix conflicts.
The algorithms that produce a mesh should ideally be changed to pass the preconstructed output mesh as a parameter, rather than internally allocating a new BufferMesh and forcing that as the output type.
I have opinions about the package structure and class naming, but it can wait till after we merge (but before we release).

src/main/java/net/imglib2/mesh/alg/hull/ConvexHull.java

ctrueden · 2023-09-11T23:37:23Z

src/main/java/net/imglib2/mesh/alg/hull/ConvexHull.java

+		final List< TriangularFacet > facets = new ArrayList<>();
+		final List< TriangularFacet > facetsWithPointInFront = new ArrayList<>();
+
+		epsilon = computeHull( vertices, facets, facetsWithPointInFront );


http://tom7.org/papers/epsilon.pdf 😉

src/main/java/net/imglib2/mesh/Meshes.java

src/main/java/net/imglib2/mesh/MeshShapeDescriptors.java

pom.xml

src/test/java/net/imglib2/mesh/alg/StupidMesh.java

ctrueden · 2023-09-12T01:03:54Z

src/main/java/net/imglib2/mesh/Meshes.java

@@ -375,4 +376,58 @@ public static void scale( final Mesh mesh, final double[] scale )
 			vertices.set( i, x * scale[ 0 ], y * scale[ 1 ], z * scale[ 2 ] );
 		}
 	}
+
+	public static BufferMesh merge( final Iterable< Mesh > meshes )


I would suggest to accept a Mesh out as argument, and concatenate them all into that, rather than forcing the result to be BufferMesh.

src/main/java/net/imglib2/mesh/Meshes.java

Original author: Tim-Oliver Buchholz (University of Konstanz) I removed the parts specific to imagej-ops infrastructure, and made it a static method. I also imported and adapted the JUnit tests from Tim.

From Tim-Oliver's code.

Now that I can compute stuff separately on MATLAB, I can add my own tests as well.

Adapted from a MATLAB code: https://fr.mathworks.com/matlabcentral/fileexchange/48913-rigid-body-parameters-of-closed-surface-meshes

The test fails in my case. And I could not understand how in Tim-O calculations we go from the inertia matrix to the ellipsoid semi- axes. BoneJ has another definition. Both give different results from what is expected.

With the BVV and a spinner to control smoothing.

net.imglib2.mesh.obj → net.imglib2.mesh.imp net.imglib2.mesh.obj.Mesh → net.imglib2.mesh.Mesh (also interfaces for triangles and vertices)

net.imglib2.mesh.obj.transform → net.imglib2.mesh.view

…util classes.

- private no arg constructor - better javadoc - remove unneeded casts to int.

tinevez · 2023-09-13T13:03:00Z

thank you @ctrueden !
I implemented most of your suggestions.

I could not find a good way to change the design as you suggested, that is: have method that produce a new mesh accepts a pre-allocated Mesh object that they mutate. The main mesh implementation we use is the BufferMesh (e.g. in the BVV), which requires the number of vertices and triangles to be stored to be known in advance.

tinevez · 2023-09-13T13:03:24Z

Also: careful I git-push-forced.

ctrueden requested changes Sep 12, 2023

View reviewed changes

ctrueden mentioned this pull request Sep 12, 2023

Are the dependencies really OK? #4

Closed

tinevez added 28 commits September 13, 2023 14:24

Import the default 3D convex hull implementation of imagej-ops.

9b6c53f

Original author: Tim-Oliver Buchholz (University of Konstanz) I removed the parts specific to imagej-ops infrastructure, and made it a static method. I also imported and adapted the JUnit tests from Tim.

Add the surface area computation to the Meshes gateway.

3959df8

Static utilities that compute various shape descriptors of a mesh.

02df625

Move the 'volume' method to MeshShapeDescriptors

373350d

Move test resources to the right folder.

a9e0079

Fix convexity computation.

f945803

WIP: JUnit tests for morphological descriptors.

6571c44

Import compactness.

4e9e6b1

From Tim-Oliver's code.

Import mesh centroid code (center of fravity for the volume).

0f91b88

Import intertia tensor code.

301d525

Mesh sparseness.

4fd821b

Code style changes.

08124cb

JUnit test for mesh volume.

b6bc6ea

Now that I can compute stuff separately on MATLAB, I can add my own tests as well.

Rewrite the centroid formula.

26c91d6

Adapted from a MATLAB code: https://fr.mathworks.com/matlabcentral/fileexchange/48913-rigid-body-parameters-of-closed-surface-meshes

JUnit test for the centroid.

0596bfb

JUnit test for the inertia tensor values.

48e8c62

Remove sparseness computation and test.

7a2958a

The test fails in my case. And I could not understand how in Tim-O calculations we go from the inertia matrix to the ellipsoid semi- axes. BoneJ has another definition. Both give different results from what is expected.

Add the icosahedron utility.

99fc040

Add an ellipsoid fitter.

ed342a1

Remove tests for unimplemented methods.

81296db

Fix minor eclipse warnings.

b1f74dc

Specify that removeDuplicateVertices returns a BufferMesh.

9d38aa4

Add BVV to test scope.

0fb6a72

Interactive demo of the Taubin smoothing.

ca09fcd

With the BVV and a spinner to control smoothing.

Fix path error in JUnit test.

9b6716c

A method to merge several meshes into one.

85f99ed

Fix change of size() method name in a test.

a6f37b7

Move core interface in the base package, rename obj package.

bc4f843

net.imglib2.mesh.obj → net.imglib2.mesh.imp net.imglib2.mesh.obj.Mesh → net.imglib2.mesh.Mesh (also interfaces for triangles and vertices)

tinevez added 10 commits September 13, 2023 14:32

Move the transform related classes.

0bee399

net.imglib2.mesh.obj.transform → net.imglib2.mesh.view

Move the test classes also in the impl package.

f4b0315

Rename MeshShapeDescriptors -> MeshStats

28e67be

Remove the interactive Taubin smoothing test and associated dependency.

74de315

Remove unneeded imports.

1a9a443

Remove unneeded cast.

83110c6

Add a private no-args constructor to prevent instantiation of static …

6ee354f

…util classes.

Fix some comments and javadoc typo. Use a DOI in a href.

2aabee0

Tweak Icosahedron class.

6d99bbe

- private no arg constructor - better javadoc - remove unneeded casts to int.

Rename EllipsoidFit -> Ellipsoid

1f9e97d

tinevez force-pushed the import-ops branch from a0290d1 to 1f9e97d Compare September 13, 2023 13:01

tinevez merged commit 111420f into master Sep 18, 2023
1 check passed

tinevez deleted the import-ops branch September 18, 2023 15:15

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Import the algorithms present in imagej-ops #1

Import the algorithms present in imagej-ops #1

tinevez commented Jul 3, 2023

ctrueden left a comment

ctrueden Sep 11, 2023

ctrueden Sep 12, 2023

tinevez commented Sep 13, 2023

tinevez commented Sep 13, 2023

Import the algorithms present in imagej-ops #1

Import the algorithms present in imagej-ops #1

Conversation

tinevez commented Jul 3, 2023

ctrueden left a comment

Choose a reason for hiding this comment

ctrueden Sep 11, 2023

Choose a reason for hiding this comment

ctrueden Sep 12, 2023

Choose a reason for hiding this comment

tinevez commented Sep 13, 2023

tinevez commented Sep 13, 2023