v0.0.6

What's Changed

• Add apps.nlinv_pics: self-calibrated wavelet SENSE reconstruction by @mcencini in #117
• Split apps.nlinv_pics into nlinv_maps, pics and partial_fourier by @mcencini in #118
• tools.whiten returns BART's whitening matrix and noise covariance on request by @mcencini in #119

Full Changelog: v0.0.5...v0.0.6

v0.0.5

What's Changed

• Public API cleanup, mr* utilities in bartorch.tools, apps on the command line, and a docs overhaul by @mcencini in #106
• Deep learning building blocks: networks for complex volumes, patchwise execution, staged and self-supervised training, uncertainty by @mcencini in #107
• Learning lessons: stated aims, MRI terminology, and figures that show the difference by @mcencini in #108
• Docs: dark-mode figures, flatter navigation, MR prose, and reworked examples by @mcencini in #109
• io: ISMRMRD raw data, DICOM and NIfTI images as tensors by @mcencini in #110
• Flat public namespaces; user guide for installation and support only; gallery nested by section by @mcencini in #111
• Fit mobafit's water/fat, diffusion and Z-spectrum models with TorchSim by @mcencini in #112
• Gallery: readable figures, and reconstructions that make their point by @mcencini in #113
• TorchSim's Bloch and EPG sequences pinned to BART's sim and epg; moba -P, -G, -D and --bloch through the app by @mcencini in #114
• ecalib takes one slice without a z axis; tools names where a missing BART command lives by @mcencini in #115
• Docs: From BART and sigpy by @mcencini in #116

Full Changelog: v0.0.4...v0.0.5

v0.0.4

What's Changed

• Fix the macOS wheel's first NUFFT, report coverage to Codecov, add badges by @mcencini in #94
• Point every link at pulserver/bartorch by @mcencini in #95
• Build BART from the downstream fork pulserver/bart by @mcencini in #97
• Fixed-width ABI over BART's 64-bit types, and a Windows build by @mcencini in #98
• CI: GCC 14 on Linux, and the CUDA wheel loaded from every Python by @mcencini in #99
• Build FINUFFT and cuFINUFFT from an upstream submodule by @mcencini in #100
• Let a source build take FINUFFT's FFT from oneMKL; keep DUCC0 in the wheels by @mcencini in #101
• Build FINUFFT for x86-64-v2, v3 and v4 as well, and pick the level at run time by @mcencini in #102
• Use oneMKL's FFT in FINUFFT when the mkl extra is installed by @mcencini in #103
• Move BART to Codeberg master with the rebased downstream stack by @mcencini in #104
• Point the BART fork page at the fork's sync workflow by @mcencini in #105

Full Changelog: v0.0.3...v0.0.4

Bartorch v0.0.3post

What's Changed

• Fix the macOS wheel's first NUFFT, report coverage to Codecov, add badges by @mcencini in #94

Full Changelog: v0.0.3...v0.0.3post

Bartorch v0.0.3

Full Changelog: v0.0.3...v0.0.3

Bartorch v0.0.2

What's Changed

• Fix the release pipeline, and let it be run without spending a tag by @mcencini in #17
• Stop building a 32-bit Linux wheel nobody could install by @mcencini in #18

Full Changelog: v0.0.1...v0.0.2

Bartorch v0.0.1

What's Changed

• bartorch: replace legacy SWIG bartpy with PyTorch-native BART interface by @mcencini with @Copilot in #1
• docs + refactor: fix CI docs build, revise architecture plan, implement base bartorch layout by @mcencini with @Copilot in #2
• Restructure bartorch: CLI tools to tools/, typed auto-generated wrappers from BART source, axes API, full pics flags by @mcencini with @Copilot in #3
• Refactor bartorch: remove pipe, consolidate tools into _generated.py, introduce _commands.py override layer, improve API and tests by @mcencini with @Copilot in #4
• Remove _dispatch.py, migrate to src/ layout, fix docs/lint/CI, embed BART with MKL-linked BLAS+FFT by @mcencini with @Copilot in #5
• feat: FINUFFT silent grid2/grid2H + rolloff replacement via --wrap linker substitution by @mcencini with @Copilot in #6
• feat: torch prior plug-and-play via --wrap nlop_tf_create by @mcencini with @Copilot in #7
• fix: fix BART FFT correctness using MKL DFTI shim; fix ecalib segfault; fix scale API; make axis/bitmask API private by @mcencini with @Copilot in #8
• Rewrite test suite: faithful BART CLI ports, fix broken API calls, add macOS CI by @mcencini with @Copilot in #9
• feat: CUDA support — zero-copy GPU tensors, cuFINUFFT GPU gridder, auto CUDA lib detection from pip by @mcencini with @Copilot in #10
• feat: expose persistent BART library operators, CG solver, deepinverse interop, and NUFFT plan caching by @mcencini with @Copilot in #11
• Reorganize package structure: split ops into linop/nlop, add catalogue by @mcencini in #13
• Restructure the API sigpy-style and rebuild the docs on autodoc by @mcencini in #14
• Take axes, index sets and prox terms wherever BART takes bitmasks by @mcencini in #15
• Follow the repository to its new name and its new default branch by @mcencini in #16

New Contributors

• @mcencini with @Copilot made their first contribution in #1
• @mcencini made their first contribution in #13

Full Changelog: https://github.com/mcencini/bartorch/commits/v0.0.1