56bcc62e49
Decode the WHUCAD h5 vector sequences (256-level quantized CATIA/DeepCAD
command matrices) and translate them into SimpleCADAPI FTC sources,
mirroring the histjson tool suite:
- whucad_vec.py self-contained decoder (numpy only; differential-
checked against the original cadlib, 1419/1419 exact)
- whucad_select.py Topo/Select history references -> ql predicates
- whucad_to_ftc.py translator (keyword-form call emission, histjson
convention; unsupported forms annotated, never
approximated)
- whucad_profile.py feature/curve/select-form distribution + coverage
- whucad_validate.py sampled build + BRep-truth volume reconciliation
(scale-corrected: vectors carry no absolute scale)
- whucad_batch.py whole-dataset batch entry point
- check_port.py differential check vs the original cadlib decode
- test_whucad_translation_output.py output-contract tests (embedded
vectors; no dataset download needed)
Full-dataset run: 146,331 models -> 146,331 .py, 424,947 features decoded,
380,074 feature blocks emitted (89.4%). pyproject gains a 'whucad' extra
(h5py).