"Instance" plus "meshopt" seems to accidentally delete instanced meshes when used with NodeJS and ThreeJS

[kitaedesigns]

"Instance" command seems to delete meshes sometimes where post optimization the instanced assets will disappear from the model. The file in question works with gltf.report but not with gltf-transform being used in three.js via WebIO. It's strange because other assets coming from Blender seem to be fine and utilize mesh_gpu_instancing correctly. This particular asset was generated with Unreal, perhaps something in their node hierarchy is messing things up. I've provided a link to the asset below. Removing Instance() from my code has the asset appear correctly.

ASSET
https://www.dropbox.com/scl/fi/fxv1ipfo9ujwgqjim3odf/TrackerlessStudio.glb?rlkey=2uerxpwfzxw8equyqorauxi1d&st=fih69joj&dl=0

• Version: [ gltf-transform v4.1.1]
• Environment: [ Browser, Node.js, Three.js ]

CODE

import { Document, WebIO } from '@gltf-transform/core';
import { reorder, quantize, meshopt, dedup, draco, flatten, instance, join, palette, prune, resample, simplify, sparse, textureCompress, weld, unlit, metalRough } from '@gltf-transform/functions';
import {KHRDracoMeshCompression, ALL_EXTENSIONS } from '@gltf-transform/extensions'
import draco3d from 'draco3dgltf';
import { EXTMeshoptCompression } from '@gltf-transform/extensions';
import { MeshoptDecoder, MeshoptEncoder, MeshoptSimplifier } from 'meshoptimizer';


// Read the GLB file using glTF-Transform's WebIO
const io = new WebIO({credentials: 'include'})
	.registerExtensions(ALL_EXTENSIONS)
	.registerDependencies({
		'meshopt.decoder': MeshoptDecoder,
		'meshopt.encoder': MeshoptEncoder,
		'draco3d.decoder': await draco3d.createDecoderModule(), // Optional.
		'draco3d.encoder': await draco3d.createEncoderModule(), // Optional.
	});

const file_in = await io.readBinary(uint8Array);  // Uint8Array → Document

file_in.createExtension(EXTMeshoptCompression)
	.setRequired(true)
	.setEncoderOptions({ method: EXTMeshoptCompression.EncoderMethod.FILTER });

await file_in.transform(
	// Remove duplicate meshes, materials, textures, etc.
	dedup(),
	// instance({ min: 5 }),
	palette({ min: 5 }),
	flatten(),
	join(), 
	weld(),

	resample(), 
	prune(),
	sparse({ ratio: 0.2 }), 
	meshopt({encoder: MeshoptEncoder, level: 'high'}),
	reorder({encoder: MeshoptEncoder}),
	quantize()
);

[kitaedesigns]

Tested this with the CLI and got the same result, might have something to do with 'meshopt' and the other transforms happening in the same document.

[kitaedesigns]

Tested with gltf.report again running all optimize commands and then exporting with Meshopt and it produces the same error.

[kitaedesigns]

I can't get it to happen with an asset exported out of Blender that has instanced meshes, so I think it may just be a bug with Unreal's GLTF Exporter.

[donmccurdy]

@kitaedesigns thanks! Fix in progress:

• fix(functions): Ensure quantize() creates required instance TRS attributes for EXT_mesh_gpu_instancing #1586

Possible workarounds in the meantime would be either of:

• (A) skip the 'instance' step
• (B) skip compression or use Draco instead

[kitaedesigns]

Thanks @donmccurdy !

[kitaedesigns]

Hey @donmccurdy if you need another file for testing, I have this one for you, made with Blender Geometry Nodes using mesh_gpu_instancing.

https://www.dropbox.com/scl/fi/eugpyvuv2zcecwsb3krgd/bddme_GN_v3_raw.glb?rlkey=qmnclaisbh35tugh6azfmcfl3&st=ht1brle5&dl=0