51 lines (28 loc) · 1.96 KB

Changelog

All notable changes to this project are documented in this file.

2026-02-23 - fireants 1.4.0, fireants_fused_ops 1.2.0 - changes

added genericlabel support!!!

2026-02-21 - fireants 1.3.2, fireants_fused_ops 1.1.0 - changes

add masked_fusedcc to test

2026-02-21 - fireants 1.3.1, fireants_fused_ops 1.1.0 - changes

added quarterion-based rigid registration
kept legacy rigid for "legacy reasons"

2026-02-20 - fireants 1.3.0, fireants_fused_ops 1.1.0 - changes

added new functionality (2d affine registration of binary shapes using subspace based contour matching: useful for stuff like histology to MRI pre-registration
makefile with scripts to deploy quicker

2026-02-12 - fireants 1.2.0, fireants_fused_ops 1.1.0 - changes

added grid_sample and warp_composer ops for 2d
ran unittests in envs with and without fused_ops

2026-02-11 - fireants 1.1.2, fireants_fused_ops 1.0.0 - changes

added 2d cuda kernels of fused ops grid sampler, warp composer
minor fixes
add update_versions_and_changelog script for easier project update

Fixed

Data type mismatch: Fixed data type mismatch in several registration classes (abstract.py, distributedgreedy.py, greedy.py, syn.py). Reference: #77

Added

Masked losses: Support for region-of-interest (ROI) driven registration using masked variants of CC and MSE.
- Concatenate a mask as the last channel to fixed and moving images; the loss is computed only where the mask is non-zero.
- Use loss_type="masked_cc" or loss_type="masked_mse" to enable masked mode (no extra options required).
- Helper generate_image_mask_allones(image) for images that have no mask (use with apply_mask_to_image so both images have a mask channel).
- Gaussian smoothing in multi-resolution registration is applied only to image channels, not the mask channel, when masked mode is active.
- See How To: Masked losses in the documentation.