Proposal pull transforms and bones from accessor (partially possible through extensions)

[ParticlePeter]

Motivation: Looking at glTf from the OpneGL standpoint, in respect to instanced drawing and MultiDrawInderect, its nice to have (a huge amount of) transforms available as a binary chunk, directly upload-able to a TBO/UBO/SSBO.

Changes possible via extension:

• geometry, camera, light and skin pull data by index from an accessor
• multiple indices for instanced drawing (in shader possibly accessible by DrawID and InstanceID)
• works for global transforms, but local transforms must be resolved:

When local transformation is required (possible via extension):

• RootNode has an accessor reference, index and child nodes
• ChildNodes reference only accessor index and optional child nodes
• No other data is stored in nodes

To evaluate the hierarchy:

• apply animation to all (gloabl and local) transforms
• traverse nodes, pull local transform data by index and compute global transforms
• upload global transforms to a GPU buffer or access indexed to set a uniform variable

Required (minimal) changes to the Spec, supply alternatives via extensions (accessor and index):

• animation.channel.target make 'id' and 'path' properties optional
• skin make 'jointNames' optional

Interested in such an extension? Draft and example files can follow.

[pjcozzi]

Thanks for the input, @ParticlePeter. Can you please provide a minimal JSON example?

[ParticlePeter]

Yes, will do, need some time for this. Would be a cube mesh transformed by three matrices enough?

[pjcozzi]

Sure, thanks!

[pjcozzi]

@lexaknyazev any thoughts on this?

[ParticlePeter]

Sorry that the example never reached this place, it actually did in proposal/issue #889 which is an evolution of this one here. The flat_nodes_proposal.zip demonstrates the same concept but via spec changes.

Only addition would be that a node can have either transformation data or reference the data via accessor. In the case of global transforms meshes could directly source their data from accessors.

[pjcozzi]

This is something we can consider for the next major glTF revision, but we are not currently actively discussing this.