Get edge seams from UV islands in Geometry Nodes

Question

If I pass mesh in with a UV map, I'd like to get the edges of the existing UV islands from the UV data, in order to separate the mesh into comparable islands, pull the island edges out into a curve object, or whatever?

Basically, I'm looking for the UV equivalent of the Mesh Read -> Edge Neighbors node. But UV edges are between face corners, rather than vertexes, and thats making me a bit confused.

So far the only way I've found is to pack UVs (to prevent errors later, see below), split edges and unwrap mesh to UV map, merge by distance, and then transfer the mesh island edges (which now correspond to UV island edges) back to the original mesh. This seems gross and the performance doesn't scale well with facecount, and transferring it back to the original mesh was a real pain.

• The UV pack has to be done to prevent errors if the original UV map has overlapping UVs, like if you mirror an object without mirroring the UVs, or if you include 16 spikes with the same geometry and same UV map on your spiky thing.

Is there a method that just looks at the UV map, without all this jumping through geometry-transformation hoops?

Answer 1

Naive approach

Alright, let's start with your solution (it would be nice if you presented it in your question, node trees are always a good conversation starter):

I didn't put the UV (re-)unwrapping there, as it is just one of many workarounds around the shortcomings of this solution. This is the effect for Susanne: (keep in mind I've marked the seams red just for this GIF, the seam info is never accessed by geonodes; also I marked the seams manually and didn't unwrap with them afterwards, so there could be some inconsistency due to that…)

OK, time to do it the proper way…

Read Seams

If you have seams on your object and you want to read them, simply use this solution:

Geometry Nodes: set sharp or crease attribute for an edge

Except in p. 6. Select UV Seam.

Recover Seams

If you no longer have the seams, then unfortunately some important data is lost. You can't fully recover the seam data just based on geometry + corner positions in UV space, because nothing forbids you to position two sides of the seam on exactly the same spot (as if there was no seam). The below solution should detect the seam in every other case: if an edge has a mismatch between the corners of its sides, it must have been a seam! The solution, then, comes down to implementing a topology node Corners of Edge, as somehow it is missing:

So let's first create a mapping that allows us to specify not just a face and a sorting weight, but (face, edge, vertex) tuple, which uniquely identifies each corner. Since each corner has exactly one associated vertex, and exactly associated face, the only problem to solve is that each corner has two associated edges - therefore I duplicate the geometry convert corners to points (because corners can't exist outside of faces and can't be found using Sample Nearest node), and set both branches' positions to $x=$ corner's face index, $z=$ vertex index, on one branch set $y=$ "previous" edge, and on the other $y=$ "next" edge then join those branches together. Now each corner has become a point, has retained it's UVMap attribute, and is positioned based on the indices of its face, edge and vertex:

Eh, I'm such a graphoman, so much text for just 9 nodes… This positioning on indices (typically with just a single index triplicated upon a conversion to a vector) is a common pattern in geonodes and shouldn't require that much explanation…

All we need to do now is for each edge, is to construct this (Face, Edge, Vert) tuple, by taking its 1st vertex index (easy, Edge Vertices node), self index (even easier, Index node), and 1st face index. Read from the mapping the corner associated with this tuple, and then do the same with the same tuple but with 2nd face index - 1st and 2nd face indices are the indices of the neighboring faces of the edge. If the UVs match - repeat the same, but this time with the 2nd vertex of the edge. If the UVs match again - it's NOT a seam:

There's just one problem… WHERE'S THE Faces of Edge NODE??

Again, need a mapping… That part I've already done here:

geometry nodes - Get Faces of Edge

The whole setup then becomes (together with the "Naive approach" for comparison):

Blender 4

Blender 4.0 is currently in βeta, meaning it can be buggy but its feature set should remain unchanged, including the Corners of Edge node. This means people coming here in future shouldn't have problems using this setup in Blender 4.0.0: