0.6.2

Maintenance release.

• [losses] Fix MI loss to allow sample ratio of 1 (#139)
• [doc] Fix regularization weights of pairwise registration tutorial (#142)
• [ci] Update used GitHub actions to latest versions using Node.js 20 (#143)

0.6.1

Maintenance release with some minor breaking changes.

• [core] Fix spatial derivatives when using mode="bspline" or mode="gaussian" (#135).
• [core] Rename SpatialDerivativeKeys all() and unmixed() argument ndim to spatial_dims (#132).
• [data] Fix exception when passing a FlowFields batch tensor to F.grid_sample() (#134).
• [losses] Add spacing option to flow field loss modules (#136).
• [spatial] Reverse stride order when subsampling DenseVectorFieldTransform (#133).

0.6.0

Add low-level functions for stationary velocity field based diffeomorphic spatial transforms.

• Modify mi_loss() to allow use of num_samples to limit maximum number when using a sample_ratio (#121).
• Ensure Gaussian kernel tensor is 1-dimensional even when sigma=0 (#122).
• Fix compose_flows() to compute composite displacement field (#128, #129).
• Add compose_svfs() to compose velocity fields using Baker-Campbell-Hausdorff (BCH) formula (#129, #130).
• Add lie_bracket() of vector fields which is required for BCH formula (#129).
• Add logv(), the inverse map of expv(), based on the algorithm proposed by Bossa & Olsom (2008) (#131).
• Switch from .vscode/ folder to deepali.code-workspace file (#123).

0.5.2

Maintenance release.

• Fix calculation of flow field Jacobian determinant (#120).
• Add add_identity option to jacobian_det() (#120).

0.5.1

Maintenance release.

• [core] Ignore error in deepali.core.pathlib.delete() of failed deletion of non-empty directory
• [utils] Fix check if path exists in deepali.utils.simpleitk.imageio.read_image()
• [utils] Force pull file from cloud storage in deepali.utils.vtk.polydata.read_polydata()
• [ci] Update action/checkout to v4

0.5.0

Maintenance release with some minor breaking changes.

• Fix DataclassConfig.from_path() when argument is a URI (#111)
• Fix and extend flow field derivative based losses (#117)
  • Add flow_derivatives() with default spacing for finite differences assuming a normalized flow field.
  • Add curl(), divergence(), and divergence_free_flow() functions to core library.
  • Modify finite_differences() to pad input using replicate instead of padding output with zeros.
  • Add mode='forward_central_backward' to avoid padding using replicate or constant values.
  • Add mode='bspline' to spatial_derivatives().
  • Generalize cubic_bspline_jacobian_*() to jacobian_*() functions for all mode values.
  • Fix bending energy, curvature, and divergence loss functions.

0.4.3

Maintenance release.

• Move deepali.utils.storage module to deepali.core.
• Add EulerRotation.matrix_() setter.
• Add LogLevel convenience initialization methods.
• Fix setting default log_level in configure_logging().
• Fix flattening of direction cosines matrix in write_sitk_iamge().
• Remove pyyaml dependency, only use ruamel.yaml instead.
• Install conda-devenv and conda-lock in conda environment.

0.4.2

Maintenance release.

• Modified spatial.ImageTransformer to allow target grid domain to differ from transform.grid() domain.
• Change image tensor creation functions to return non-batched image tensor when num=0 is passed.
• Fix assertion condition when comparing data tensor shape to MetaImage DimSize metadata.
• Fix torch.from_numpy() warning that NumPy array is read-only when reading MetaImage from data blob.
• Fix circle_image() and cshape_image() functions used to create synthetic 2D registration example images.

0.4.1

Minor refactoring of deepali.core and deepali.utils with a few breaking changes.

• Renamed deepali.core.path to deepali.core.pathlib
  • Moved functions make_temp_dir(), temp_dir(), and temp_file() to deepali.core.tempfile.
• Renamed deepali.core.types to deepali.core.typing.
  • Moved TensorCollections related functions to deepali.core.collections.
• Move deepali.utils.cli modules to deepali.core such as
  • deepali.core.argparse
  • deepali.core.environ
  • deepali.core.logging
• Added deepali.utils.imageio library.
  • If installed, use nibabel to read/write NIfTI images instead of SimpleITK.
  • Support direct streaming of .mha MetaImage files from cloud storage (AWS S3).
  • Keep using SimpleITK for all other image file formats.
• Modified Image.read() and Image.write() methods to use deepali.utils.imageio library.
• Use absolute imports between deepali.* packages, relative imports only within.
• Added software citation using Zenodo.

0.3.2

Maintenance release with fix of ImageBatch.__getitem__ handling of int index along channel dimension. Includes support of more batch tensor indexing. Added an introductory tutorial notebook to walk through an image registration with different spatial transforms.