gltfpack: Visible quantization artifacts on some scenes

[Mateo150]

I recently update to the latest version, v.0.18.
And realiced that now is very common after a simple optimization, even without compression the geometry looks "pixelated"
gltfpack -i ... -o ...

How it should look.

How it actually look.

I think this was due to some issue on my meshes, but after the update this happen almost every time.
with v.0.16 this neeeeever happen

Any idea why this is happening?

[zeux]

This looks as if something changed with quantization behavior between 0.16 and 0.18. Please attach the reproduction scene - without it it's impossible to diagnose cases like this. You can also pass -vp 16 to improve this behavior.

[zeux]

P.S. It's actually a little surprising that this scene used to work without quantization artifacts without extra options. You might want to use -kn flag because without it gltfpack will try to transform meshes to world space if they are static.

[Mateo150]

Thanks for answering!

1. Not necesary only on larger scenes. This is another one, very simple, in this case the character is rigged.
   
   A thing is that I realized, it only happens in certain meshes with the property in common than the vertex are very close one of each other. (This could be a factor?)

2. -vp 16 makes it looks better but still have trace of those squares. In v0.16 with default value works perfectly.

3. Yeah sure, in this scene I need to use -kn specialy and other flags, I mention that no extra options to make it clear that even in the simplest way happen that "squarization".

4. https://we.tl/t-pvsdQ4kGkH
   Here is this character scene without quantization, as blender exports it. And this is how I quantize it and outputs as the previous image gltfpack -c -i ./Ar_v4.glb -o ./Ar_v4c.glb

[zeux]

The scene above has the same issue on gltfpack 0.16 as far as I can tell. This came up before (#433) and is due to some odd export behavior that Blender has with skinned meshes and unit conversions, where it exports skinned geometry at a very different resolution vs non-skinned geometry.

I am not sure if this is fixable, would need to think about the consequences of propagating scale from the skinned model into the vertex data. Recently added -vpf fixes it but it's not very widely tested.

[zeux]

Note: for any report in this thread that doesn't come with a link to the scene before the compression, it's going to be impossible to investigate whether it's something that can be improved or not. So far out of 3 examples here only one (from comment #466 (comment)) has been reproduced and traced to a skinning export issue, tracked in #433.

[scarletsky]

Hi, I just have a test on a model on sketchfab
https://sketchfab.com/3d-models/fictional-supercar-v12-goblin-0a20e49ad5774d778567cb5c3f345786

I do these steps with the model:

• convert it to GLB with blender 3.2.2
• use gltfpack -cc to output a meshopt glb
• put the result to https://gltf-viewer.donmccurdy.com/ to see the result

The result is OK, and I modify the model, add a really large sphere on the model. This time, I got artifacts on gltfpack -cc, and got better result with -vp 16

OK model.glb.zip

OK model_meshopt.glb.zip

OK model_with_sphere.glb.zip

Artifacts model_with_sphere_meshopt.glb.zip

Little artifacts model_with_sphere_meshopt_vp16.glb.zip

[zeux]

Since no new examples have been posted, and #466 (comment) is inevitable / expected, I'm going to assume that there's no regression and that the only tangible issue is with the skinned models and a very large scaling factor that Blender adds, which will be tracked / fixed by #433.