v1.5

• extensively refactored code base
• self-contained, linajea_experiments repository is no longer required
• improved documentation
• new examples for improved usability
• all deep learning code switched to pytorch
• tracking code cleaned up and made modular and extensible

v1.4

Main bug fixes and new features:

• use attrs-based configuration (everywhere)
• add code to output data required for sSVM to automatically determine ILP weights
• improve handling of cell state/cycle information

(to be used in combination with the matching branch in linajea_experiments: https://github.com/funkelab/linajea_experiments/tree/v1.4-dev)

(code used for experiments in miccai2022 paper)

v1.3.1

Main bug fixes and new features:

• improve greedy tracking
• add function to evaluate percentage of perfect reconstruction
• add scripts to export to CTC format

(to be used in combination with the matching branch in linajea_experiments: https://github.com/funkelab/linajea_experiments/tree/v1.3.1)

(code used for experiments in NBT paper)

v1.3

New Features

• full frame division evaluation
• matching gt and candidate nodes to create best possible tracks for a given candidate graph
• "validation score" that encourages long, continuous tracks without punishing too much for less than single-cell localization
• option to use parent vector distance or euclidean distance when extracting edges
• ability to pass a list of parameters to solver, to allow more efficient grid search parallelization
• a greedy tracking algorithm that picks the shortest edges in a per-frame fashion
• a streamlined, minimal ILP (along with the old, non-minimal version)
• streamlined, minimal way to incorporate cell cycle predictions into the minimal ILP

Improvements

• more robust mamut export scripts

v1.2

• add code to assist with analysis of results
• add validation metric that emphasizes continuous tracks
• update gunpowder TracksSource to read tracks files with headers
• add Spimagine visualization code

v1.1.1

• add support for timelapses in zarr containers
• use faster cKDTree
• allow prediction without singularity

v1.1

Major changes:

• Upgrade tensorflow to v1.14 and cudatoolkit to v10.0
• Rewrite evaluation to evaluate divisions separately
• Bugfix: read selected attr when solving blockwise to ensure pin constraints from neighboring solutions

v1.0

Merge pull request #4 from funkelab/1.0-dev

Merge 1.0 dev branch into master for release