xatlas introduces manifold issues

[chpatrick]

I'm trying to run xatlas on this (admittedly kind of screwy) mesh. This has previously had manifoldness issues fixed, sometimes by duplicating vertices in the same position. However, after I run xatlas, I get this mesh out. According to Meshlab, it has 136 non-manifold vertices. Maybe the vertices in the same place got re-merged? Is it possible to ask xatlas to not change the geometry of the mesh at all?

I'm using

xatlas::ParameterizeOptions parameterizeOptions;
parameterizeOptions.closeHoles = false;
parameterizeOptions.fixTJunctions = false;

[jpcy]

The only change xatlas makes to geometry is adding duplicate vertices on chart seams - since those vertices now have different texture coordinates. This should always be true. Those parameterization options only affect temporary meshes.

I don't see any texture coordinates importing the output mesh into Blender.

[JulianKnodt]

Sorry to revive a dead issue, but I also ran into this. I was wondering if it would be possible to reuse 3D vertex coordinates, and duplicate only the UV coordinates.

Currently, there is a 1-1 relationship between 3D vertices and UV parameterized points in xatlas, but the OBJ format allows for specifying different UV coordinates depending on the face, with the same 3D vertex. Is there any way to enable this, or a way to add this change into xatlas? I'm willing to do this if I can get a little guidance into where it may be possible.