B1's MoCapAnything pick reversed (animal model, wrong for human capture);
license analysis recorded in LICENSES.md. John chose Path B: run HSMR->SKEL
locally on ultra (MPS), ship rig-motion only, accept the non-commercial
licence risk as interim until a commercial SKEL licence request goes out
(Path A, the clean upgrade).
Engine installed under .engine/ (gitignored): uv venv py3.10, torch 2.13
MPS-green, ungated weights, smoke_test.py with Apple-Silicon bypasses
(detectron2 stubbed with full-frame bbox, headless pyrender stubbed -> QC in
Blender) -- see docs/B_ENGINE_SETUP.md.
Remaining before B ships: John's gated SKEL v1.1.1 download, then the
forward-pass smoke test (MPS end-to-end, timing, pose sanity). Known caveat:
HSMR is per-frame so root is camera-space cam_t, NOT world-grounded -> Lane C
derives root from foot-contact.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Fresh SOTA survey (per PLAN — distrust the snapshot). Findings:
- MoCapAnything V2 (MIT code+weights, no SMPL, BVH-direct onto arbitrary
rig) is the pick — sidesteps the SMPL commercial-license hard-stop and
collapses the SMPL->mixamorig mapper + most of Lane C retarget.
- GVHMR: best accuracy but non-commercial code license -> reference only.
- GEM-X: commercial-clean + perfect output but CUDA-only -> off-fleet.
Stopped at the John-decision gate (docs/DECISIONS.md): confirm engine,
clear m3ultra disk (97% full), supply test footage. Inference eyeball,
B2 service-ize, B3 jitter remain.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
