Removing KHR_mesh_quantization

[donmccurdy]

Asked in the three.js repo, how can you remove the KHR_mesh_quantization extension from a model?

[donmccurdy]

Should be something like...

// remove_quantization.js
const { NodeIO } = require('@gltf-transform/core');
const { KHRONOS_EXTENSIONS } = require('@gltf-transform/extensions');

const io = new NodeIO().registerExtensions(KHRONOS_EXTENSIONS);
const document = io.read('input.glb');
const root = document.getRoot();
const tmpAttr = document.createAccessor('TMP');

// Replace int8/int16/uint8/uint16 attributes with float32.
for (const mesh of root.listMeshes()) {
	for (const prim of mesh.listPrimitives()) {
		for (const semantic of prim.listSemantics()) {

			const attr = prim.getAttribute(semantic);
			if (attr.getComponentSize() === 4) continue;
			if (semantic.startsWith('JOINTS_')) continue;

			// Create identical tmp attribute with float32.
			tmpAttr.copy(attr)
				.setNormalized(false)
				.setArray(new Float32Array(attr.getArray().length));

			// Copy data to tmp attribute.
			for (let i = 0; i < attr.getCount(); i++) {
				tmpAttr.setElement(i, attr.getElement(i, []));
			}

			// Merge changes back to original attribute.
			attr.copy(tmpAttr);

		}
	}
}

tmpAttr.dispose();

// Remove KHR_mesh_quantization.
root.listExtensionsUsed()
	.find((ext) => ext.extensionName === 'KHR_mesh_quantization')
	.dispose();

io.write('output.glb', document);

And then run the script with node remove_quantization.js.

If quantization was applied with certain tools, the model might have KHR_texture_transform listed under extensions with gltf-transform inspect model.glb. If this UV transform works as expected already there is no need to remove it. If you need to remove the UV transform as well, it would need to be applied to the UVs, not shown above.

If some of the attributes are normalized (this will print a warning in the code above) there may be some extra conversions needed; I'm not sure if that is commonly the case.

[carstenschwede]

Thanks for the code. I actually do have KHR_mesh_quantization, KHR_texture_transform, KHR_materials_pbrSpecularGlossiness in my particular model and I also get the warning May need to apply normalization ratio!.

The resulting model has some objects at very different scales in comparison to the original. Any idea?

[donmccurdy]

Ok – I've made some edits above to handle the normalization. KHR_texture_transform should be fine as long as your textures are appearing correctly on the model.

[carstenschwede]

Thanks a lot for the help. The output.glb model created using your script looks good (left). But exporting it using GLTFExporter and reimporting it again fails (right).

GLTFExporter shows the following warnings:

• GLTFExporter: Creating normalized normal attribute from the non-normalized one.
• GLTFExporter: Attribute "skinIndex" converted to type UNSIGNED_SHORT.

[donmccurdy]

Updated with another guess — if that's not working and you'd be able to share the model (my email address is on https://github.com/donmccurdy) I can try debugging that too. The first warning looks OK.

[carstenschwede]

Awesome, thanks a lot. Looks good now!