Retrieved CC0 rig, facial poses and body targets
The optional authored-person adapter now has verified local core data under tests/.work/anatomy-candidate/raw/. Asset manifest records the exact source URL, reconstructed byte size, SHA-256 and per-file provenance for every asset. Every collected file has complete numbered-line coverage and zero conflicting duplicate lines. No program source was copied. MakeHuman asset license.
Rig and weights
| Input | Local bytes | Validated content |
|---|---|---|
default.mhskel | 117,619 | 163 bones, 326 vertex-defined joints, 163 roll planes; no missing parents/head/tail references or parent cycles |
default_weights.mhw | 897,521 | 57,107 finite nonnegative pairs across 139 bones; all 19,158 source vertices weighted; no unknown bones, duplicate pairs or invalid vertex indices |
face-poseunits.bvh | 310,766 | 163 matching bones, 492 channels, 60 complete frames; all frame values finite |
face-poseunits.json | 1,581 | 60 named pose-unit frames; explicit CC0 metadata |
The original weights require normalization. Body vertex sums range from 0.954 to 1.063. Of the 13,380 body vertices, 2,460 have more than four influences; retaining only four can omit 30% of normalized weight. Retaining eight affects only 14 body vertices, with maximum omitted normalized weight 0.4975%. These figures describe source weights, not resulting deformation error. All original influences remain available. The official technical notes specify normalized sums and describe reference-bone retargeting. Weight and pose conventions.
The facial BVH uses root translation plus X/Y/Z rotations, and X/Y/Z rotations for the remaining bones. Its JSON names frames such as Rest, left/right upper-lid closure, cheek raising, mouth-corner movement, brow movement and jaw drop. They are individual pose units; they are not a chronological facial animation. A source-authored complete smile or surprise preset was not retrieved. The JSON's animation_file still says poseunits.bvh; the actual paired repository file is face-poseunits.bvh. Frame metadata, pose data.
An independent comparison of all 163 zero-rotation BVH joint origins against the OBJ vertex-defined rig supports the coordinate conversion (x,y,z) → (x,z,-y). Its best positional fit has a 3.30 cm RMS residual, so the reference rest shape is not identical. This is an inference from actual data, not a verified official declaration of the BVH convention. Do not substitute the fitted scale/translation for proper rest-frame retargeting. Numerical coordinate audit.
Body target set
All targets are original-index displacement data for hm08 and keep the OBJ topology unchanged. The four baseline files are caucasian-{female,male}-{young,child}.target, each containing 19,150 unique finite deltas. The legacy source category is fixed for this bounded evaluation; it does not introduce an ancestry control or claim that this category exhausts natural human variation.
The adult build family consists of universal-{female,male}-young-averagemuscle-{minweight,averageweight,maxweight}.target. Both adult average-weight files are verified empty-delta targets. The two corresponding child average-weight files are also verified empty. The eight nonempty baseline/build targets and four empty regular targets are under raw/targets/.
The official source separates the baseline macro group from the age/sex-dependent weight/muscle group. For exact presets, use one selected baseline and one matching build delta; for intermediate controls, blend the corresponding endpoints within both families. Do not stack both sex baselines at full weight, apply adult build deltas to children, or treat age-category names as verified chronological ages. This is a bounded independent composition using the documented data groups, not a copied MakeHuman slider implementation. Target organization and dependencies, target inventory.
Baseline-only body heights before requested stature scaling are 1.60542 m for female/young, 1.74813 m for male/young, 1.26718 m for female/child and 1.35650 m for male/child. These come from applying actual source deltas and measuring body bounds, not uniformly shrinking the adult model. Shape audit.
Remaining limits
The four listed universal-{female,male}-child-averagemuscle-{minweight,maxweight}.target files returned Cache miss through both the official raw and GitHub file-view routes. They were not reconstructed, approximated or replaced with adult data. The adapter must expose only validated child build options until these files can be obtained.
No individual skin UV atlas was verified or ingested. The official skin-pack inventory identifies CC0 skin assets, but linked individual asset pages refused connections or failed retrieval. The official organization lists a bundled asset repository, but its root was unavailable through the web tool and the requested master tree returned 404. Rendered head thumbnails were not substituted for UV maps. Skin material quality remains an unresolved visual requirement. Official skin inventory, official repository listing.
The static preview remains separate from the optional adapter and the current working Person. Its visual findings and unresolved texture-counter cleanup assertion are recorded in candidate validation. Posed deformation, facial expression appearance, subdivision cost and full scene integration are the adapter task's independent acceptance work.