A new fun little benchmark: A 10x10x10 grid of cubes: union them all into one solid block, then difference them all back out again. 1999 iterated ops whose correct result is... nothing (the empty set).

This Minecraft-esque case is relatively hard because the intermediate results are completely solid but not necessarily manifold (depending on your definition of choice).

#CAD #ComputationalGeometry #3D #geometry #GeometryProcessing

https://solidean.com/blog/2026/iterated-cube-grid-benchmark/

Yay another mesh boolean benchmark! This time carving a terrain from a cube using sweep-volume-spheres. With fancy videos inside (also from some failure cases with interesting patterns). https://solidean.com/blog/2026/terrain-carve-benchmark/

#CAD #ComputationalGeometry #3D #geometry #GeometryProcessing

A new twisting beam example added to FEBio.jl. This example uses 20-noded quadratic hexahedral elements. The left end of the beam is supported and rotation of the right beam is prescribed.

https://github.com/febiosoftware/FEBio.jl (see demo_febio_0011_beam_twist)

#opensource #FiniteElementAnalysis #GeometryProcessing

Field-Aligned Surface-Filling Curve via Implicit Stitching is out! 🎉

We present a robust and scalable method to generate field-aligned surface-filling curves on general manifolds, from interactive small models to massive meshes.

Project page: https://xavierchermain.github.io/publications/surface-filling-curve

We'll present it at #Eurographics2026 in Aachen this May.

Affiliations: Université de Lorraine, @cnrs @inria and @Labo_Loria

Thanks to the MFX team and its leader @sylefeb

#ComputerGraphics #GeometryProcessing

Graph and levelset based surface construction.

The animation shows a graph curve on the left that is coloured towards a desired local radius. Such a curve could stem from an image segmentation such as the centre line for a blood vessel. The local radius would then represent the local blood vessel radius. On the right a slice through a corresponding "levelset image" is shown. The level set image is a signed distance for the vessel surface but normalised by this radius, so a distance of 1.0 means we are 1 local radius value away from the desired surface. An intensity level of 0.0 means we are on the desired surface, negative levels mean are for the inside, and positive levels are for the outside. Once the levelset image is established the surface can be constructed by drawing the isosurface for the level 0.0. Just to test the algorithm, in this animation, I am creating a sinusoidal variation of the radial data. By increasing the frequency there are more and more bumps in the surface. Critically, when features on the surface become too curved or thin with respect to the voxel size, the surface may become too noisy as the coarse voxelisation starts to fail to capture the detail in the geometry.

This is the new #Julialang implementation but my old MATLAB implementation was used here:

https://doi.org/10.1016/j.jbiomech.2021.110896 (open access version: https://doi.org/10.31224/osf.io/qaujs)

and here:
https://doi.org/10.1098/rsos.242025

#opensource #geometryprocessing #Comodo

Meet "spinodoid" structures. You basically splash lots of waves in all sorts of random directions with random phases, and then you threshold the resulting mess.
These structures stem from the idea of a "spinodal decomposition" and the waves form a "Gaussian random field". The latter has been linked to/used for animal patterns (stripes etc), phase separation in chemistry/metallurgy, quantum mechanical random fields, up to cosmological structures...
But here I just use it to create stochastic lattice structures. Because my waves here have different orientations but the same frequency, they show up in a Fourier transform as a circle or sphere, which I think is just neat :)

Spinodoids are coming to Comodo very soon.

More on Spinodoids:
https://doi.org/10.1038/s41524-020-0341-6

Spinodal decomposition:
https://en.wikipedia.org/wiki/Spinodal_decomposition

Gaussian random fields:
https://en.wikipedia.org/wiki/Gaussian_random_field

#opensource #Julialang #GeometryProcessing #Lattices

Angle based surface segmentation. This animation showcases the output of Comodo's faceanglesegment function. Imported surface geometry that lacks any labelling of surface features is segmented, based on surface orientations, into separate features, e.g. a cylindrical hole, a flat face, a gear tooth flank, a propellor face, and so on. This function is very handy to assign boundary conditions to particular aspects of a model, or to segment it into separate parts for remeshing.

#opensource #Julialang #GeometryProcessing

Meet Taubin smoothing. It is a nice smoothing method that is great for smoothing voxel based meshes. So for smoothing stepped "Lego-like" meshes.

The left here shows a raw voxelised (thresholded) voxel boundary surface. On the right is the Taubin smoothed result.

Taubin smoothing works by first doing an aggressive Laplacian smoothing step (a bit like a blurring), which removes bumps etc, but which unfortunately can distort the shape too (e.g. unwanted shrinkage/growth). Next it uses an "inverse smooth" (a bit like a sharpening operation) to attempt to push back the shape a bit to avoid distortion. With the right parameter choices it can perform really well.

This animation is for a thresholded 3D image. The colors are the original voxel intensities. The resulting shape presents as a central sphere like blob contained in a dodecahedron "cage".

See also: https://doi.org/10.1145/218380.218473

#GeometryProcessing #JuliaLang #Makie #Meshing #TilingTuesday

Meschers: Geometry Processing of Impossible Objects

https://anadodik.github.io/publication/meschers/

#HackerNews #Meschers #Geometry #Processing #Impossible #Objects #GeometryProcessing #3DModeling #HackerNews