API reference¶

The reference is generated from the docstrings and organized by subpackage. Most user code needs only MIR.models (networks and SpatialTransformer), MIR.image_similarity (losses), MIR.deformation_regularizer (regularizers) and MIR.accuracy_measures (metrics).

Models¶

Registration networks, transformation models (VecInt, LDDMM), spatial transformers, configurations, backbones and heads, classical methods.

MIR.models

Model zoo and configuration helpers for MIR registration pipelines.

Losses¶

Image similarity and label overlap losses.

MIR.image_similarity

Image similarity losses and metrics used in MIR registration.

Regularizers¶

Global, spatially varying and inverse-consistency regularizers, priors.

MIR.deformation_regularizer

Regularization terms for deformation fields and velocity models.

Metrics¶

Dice and Jacobian-determinant (digital diffeomorphism) evaluation.

MIR.accuracy_measures

Accuracy metrics and Jacobian-based measures for evaluation.

Utilities and data¶

Training utilities, mixed-precision and multi-GPU helpers (MIR.utils.precision, MIR.utils.distributed), augmentation, intensity normalization, synthetic images for pretraining, statistics, Learn2Reg I/O, label references and the pretrained-weight registry.

`MIR.utils`	Utility helpers for registration, training, visualization, and IO.
`MIR.data_augmentation`	Data augmentation utilities for MIR training and preprocessing.
`MIR.intensity_normalization`	Top-level package for intensity-normalization.
`MIR.random_image_generation`	Random image generation utilities for pretraining and augmentation.
`MIR.statistical_analysis`	Statistical analysis utilities for MIR experiments and reports.
`MIR.l2r_io`	L2R image format reader/writer with lightweight compression.
`MIR.label_reference`	Reference label tables for MIR datasets.
`MIR.pretrained_wts`	Pretrained weight and dataset ID registry for MIR.