support `byteStride` for glTF files

[russaa]

currently reading glTF data that utilizes byteStride in its accessors is not supported

(for example, the avatars at Ready Player Me use byteStride, see https://models.readyplayer.me/678703d63cb298e56bf50122.glb?morphTargets=ARKit&textureAtlas=1024)

I think this would need some refactoring of the gltf-loader gltf2.py:
currently, the Accessor and BufferView are intermixed in some places which prevents a easy implementation for reading buffer-data with defined byteStride.
See for example VBOInfo, or more precisely the creation of it from GLTFAccessor.info() which tries to merge data from GLTFAccessor and GLTFBufferView, and then implements its own read() method for reading & preparing the gl buffer, see implementation of GLTFMesh.load().
I think, the reason for this was, to be able to merge multiple accessors & buffer-views, to make the rendering more efficient. But the current implementation cannot i.m.o. easily changed to also support byteStride, since an efficient implementation would require information from the accessor & buffer-view.

I think it would be more clean, to implement & use an appropriate method in GLTFAccessor that supports byteStride.

As an example for how to read data with byteStride, see e.g. trimesh's implementation

I did try out the trimesh implementation and got it to work with gltf2.py, but as I mentioned, without some refactoring, the code is quite hacky.

[einarf]

Thanks for pointing out this flaw. It's definitely a bit hacky and needs improvements. It was really something I threw together over the weekend with the goal of supporting most of the files in the glTF sample repository.

[russaa]

ah, sorry for the misunderstanding:
with hacky I ment my own code after making it work -- otherwise I would have made a PR.

If you think, it could be helpful, I could still make a PR with my hacky code (?)

I think the existing code tries to load the interleaved data (with byteStride) efficiently in a single block, configuring the VBO accordingly.
But as far as I can see, the detecting & merging of interleaved data does currently not (always) work.

The implementation of trimesh de-interleaves the data, so it would be a little bit more inefficient -- but the implementation is less complicated, aside from the fact that the current structure in gltf2.py needs to be changed in a few places to make it work.

[einarf]

Definitely make a PR with your code regardless of working state. At the very least it's helpful to understand the problem.

.. and yes I think you are right about the interleaving. I remember taking some shortcuts there.

[russaa]

I added my implementation (see PR #215):

maybe, for now, the de-interleaving strategy could be used as a fallback, e.g. if merging fails, but there is a byteStride defined, load the buffer de-interleaved