simplify: how to correctly use attribute_weights and meshopt_SimplifyErrorAbsolute

[kzhsw]

I'm using meshoptimizer simplifier to simplify meshes from gltf models, and by using meshopt_SimplifyErrorAbsolute, I multiplied target_error with the maximum value of node's world scaling for each node, to make even target_error across all meshes, but when using meshopt_simplifyWithAttributes, should I use fixed attribute_weights for all meshes, or should it be multplied with node scaling, like target_error?

[zeux]

The attribute weights are implicitly taking the distance into account; from documentation:

internally, a change of 1/weight in attribute value over a distance d is approximately equivalent to a change of d in position

Thus it should be correct to use fixed attribute weights that do not depend on the mesh extents; this is what gltfpack does for multi mesh simplification as well.

[kzhsw]

Thank you for this quick answer.
I saw that in gltfpack it uses fixed weights, but after a quick look into the code, seems it uses the default relative error mode.
I saw in previous discussion using meshopt_SimplifyErrorAbsolute changes the "scaling" of position, so if changing weight has the similar effect of a diff in position, does it mean that using meshopt_SimplifyErrorAbsolute will need to scale the weights when I scale target_error based on node transfrom, or is it ok to just keep the fixed value?

[zeux]

It should be ok to keep the fixed value regardless. meshopt_SimplifyErrorAbsolute changes the scaling of the errors - target_error and result_error - but the internal calculations that prioritize collapses stay the same, and the relation between attributes and positions doesn't change. The statement "if the mesh is 50 units across, an error 0.01 without using this flag is equivalent to error 0.5 when using this flag" from the linked discussion should hold even if you add attributes (with fixed weights).

gltfpack indeed uses the default mode but that's mostly because the command line input -se that controls simplification quality is specified in relative units, not in scene units like centimeters.