Suggestion: Sample with discrete/constant interpolation of scale 0↔1

[donmccurdy]

In a thread on the three.js forums we've come to the conclusion that a significant number of viewers cannot display constant interpolation of scale, particularly scale 0↔1.

• ✅ three.js
• ✅ PlayCanvas
• ❌ Babylon.js
• ❌ Khronos Sample Viewer
• ❌ Spark AR

The 'broken' three each show slightly different results — Spark AR doesn't appear to support constant interpolation at all, whereas Babylon.js plays the animation correctly once and then reaches a broken state when trying to loop it.

It's a useful technique for 'baking' simulations as flipbook-style animations, showing each mesh for a single frame, but does depend on correct support for animating scale with constant interpolation.

[DRx3D]

This has been untouched for over a year. Is there anything to do here?

If there is no response, then this issue will be closed by 27 Nov.

[bghgary]

Babylon.js has already fixed this, but a sample model could still be useful.

[javagl]

By "constant", you mean STEP interpolation?

There is a pending PR for the InterpolationTest at #26 . I could just change the values of the respective accessor...

  accessor10.setArray(new Float32Array([
    1, 1, 1, 
    0.5, 0.5, 0.5,  // Change to 0 0 0
    1, 1, 1, 
    0.5, 0.5, 0.5,  // Change to 0 0 0
    1, 1, 1
  ]));

if this is it...

[donmccurdy]

Yes, thanks – STEP interpolation with scale is what I had in mind.

[javagl]

This is what the InterpolationTest could look like with a scale of 0:

An archive with the GLB and the code (hand-edited output of gltfTransformifier) for generating it:

InterpolationTestFix0004_scale0.zip

If people agree, I can update #26 with this.

[donmccurdy]

An archive with the GLB and the code (hand-edited output of gltfTransformifier) for generating it:

total aside, but that is a wild and beautiful solution 🤯

[javagl]

OT: The name was inspired by BCELifier. It's a very rough "proof-of-concept" for now, but I try to allocate some time to maybe turn it into something useful.

[javagl]

I have updated the PR at #26 (comment) with the latest state that was posted here, where the scale interpolations go down to (0,0,0). When this PR is merged, then this issue can probably be closed.

[javagl]

The current version of the InterpolationTest at https://github.com/KhronosGroup/glTF-Sample-Assets/tree/main/Models/InterpolationTest now includes a STEP interpolation that scales between 0.0 and 1.0, so I assume that this can be closed.