Confusing simplify results

[slimbuck]

Hello!

I'm using meshopt in splat-transform to simplify the naive marching cubes output for purposes of collision mesh generation.

I am getting slightly odd results and wonder if I'm doing something wrong?

The code I use is:

    const [simplifiedIndices] = MeshoptSimplifier.simplify(
        mesh.indices,
        mesh.positions,
        3,
        0,
        maxError,
        ['ErrorAbsolute']
    );

The resulting mesh is in places perfectly optimised and in others not so much:

Perhaps I'm using the API incorrectly?

A copy of the output mesh can be found here.

Thanks in advance!

[zeux]

The GLB file seems to simplify fine in my viewer:

Since I don't have the original mesh I don't know what issue that you're experiencing might be; the above is obviously simplifying from an already simplified mesh. But I will hypothesize that your maxError is very low and there's some floating point induced difference between the different walls that causes one wall to have an error closer to 0 than the other. You could check if enabling "Regularize" flag changes this; note that this requires increasing the error threshold a bit too.

[slimbuck]

Hi @zeux,

Thanks for the reply.

Unfortunately I just could not get meshoptimizer to reduce marching cubes output by removing only unnecessary flat verts/faces and leaving all structural detail.

It would always start removing structural detail while large flat triangulated areas remained. (I tried adding vertex normals to help with simplification, but then ran into memory issues with the wasm build).

In the end I got claude to write a topology-preserving vertex removal. Here is an example output:

The results are good, but I would much rather depend on a battle-tested solution like meshoptimizer.

Anyway, if you're ever able to add simplification like this I would take it in a heartbeat :)