Skin vertex weight normalization

[ziriax]

The spec says:

The joint weights for each vertex must be non-negative, and normalized to have a linear sum of 1.0. No joint may have more than one non-zero weight for a given vertex.

The validator seems to check if the weights sum to one in groups of 4

Is this the desired behavior?

E.g. all of my customer's models have 8 weights. If feels more logical to me that the sum of these 8 weights should be one?

I'm trying to fix Maya2glTF to get rid of the normalization errors like these:

Weights accessor elements (at indices 2744..2747) have non-normalized sum: 0.9999983310699463

[lexaknyazev]

The weights sum is accumulated across all weights sets (4, 8, 12, etc). The error message may be better, though.

Could you please provide an example asset?

[elalish]

Here is an example (and old, non-normalized version of CesiumMan) that passes validation but in fact has non-normalized weights: https://github.com/google/model-viewer/issues/2450#issuecomment-902428164

[lexaknyazev]

Both models from that archive have unorm8 skinning weights. The CesiumMan.quantize-hi.glb shows 320 relevant validation errors.

[elalish]

yeah, the quantized one is being caught just fine. But the base one just has a single warning, yet is still invalid.

[elalish]

Hmm wait, unorm8 skinning weights; that sounds like they just need to be divided by 255, yes? Maybe that is our bug; thanks!