v0.3.2

New Backends

• JAX backend (smplfitter.jax): BodyModel, BodyFitter, BodyConverter with full feature parity to PyTorch
• Numba backend (smplfitter.nb): BodyModel, BodyFitter, BodyConverter — JIT-compiled NumPy for fast CPU fitting without GPU frameworks

New Features

• BodyConverter added to NumPy and TensorFlow backends (previously PyTorch-only)
• TF BodyFitter rewritten with full feature parity to PyTorch (kid fitting, scale estimation, joint fitting, weighted fitting, etc.)
• Body model downloader (python -m smplfitter.download): interactive download of SMPL/SMPL-X/SMPL+H/MANO models from MPI servers
• Better error messages: helpful guidance when model files are not found, including env var and download instructions
• SMPLFITTER_BODY_MODELS env var: new dedicated env var for body model path (in addition to DATA_ROOT)
• rototranslate API: t parameter now defaults to zero for pure-rotation use case
• Experimental PyTorch variants: bodyfitter_opt.py and bodyflipper_opt.py with optimization-based fitting

Documentation

• Sphinx docs with pydata theme, autoapi, and intersphinx
• Improved Sphinx cross-reference resolution and source links
• How-to guide with usage examples including body model conversion
• Benchmark suite with plotting for cross-backend performance comparison

Infrastructure

• CI with ruff linting (GitHub Actions)
• PyPI publishing via trusted publishing
• Pre-commit hooks, editorconfig, dependabot
• Shared test infrastructure with conftest.py and parameterized cross-backend tests
• setuptools-scm for version management