- Factored out model specific updates from Bayesian update code
- Added features to objects
- Added visual update method
- Add deprecation warning to track_interactive
- Revert to python 3.7+ for compatibility
- Remove python keywords from codebase
- Add ability to save tracks directly to h5 files
- Allow reading/writing of nD features to h5 files
- Added util to relabel masks using track ID
- Added extra test coverage
- Changed to static setup.cfg
- Fix inclusion of napari.yaml in wheel
- Update CI checks
- Allow Q matrix to be directly defined in config file
- Use pydantic for tracker configuration
- Allow config export
- Add imaging volume dimensionality inference
- Fix warnings for implicit type conversion
- Removed _PyTrackObjectFactory
- Fix logging errors
- Add napari-hub manifest
- Add napari-btrack package
- Refactor localization to remove dask dependency
- Increase test coverage
- Bug fix to remove indexing error for objects in the last frame
- Deprecate render package
- Added dask array input capability for _localisation.py
- Updated models to use dataclasses
- Added more documentation, readthedocs page
- Added example datasets using pooch
- Replaced numpy matrices with arrays
- Renamed logger instances
- Fixed export delegator
- Allow H5 writer to overwrite existing properties
- Fixed version checker
- Bug fix for object properties. These are now updated rather than being overwritten
- Bug fix for writing multiple object types to the same h5 file
- Support for filtering by properties in h5 file
- Added project logo
- Major update of docstrings to follow numpy guidelines
- Added a separate configuration guide document to the repo
- Apple M1 support for compilation
- Added ability to replace NaN in track properties with interpolated values for visualization
- Added track and object properties which store associated metadata
- Added more tests
- Added a segmentation to objects utility function
- Allow properties to be determined directly using scikit-image
- HDF5FileHandler can now write objects directly
- Use long C type for object ID
- Refactored HDF5FileHandler for improved loading/saving
- Added refs property to Tracklets to enable indexing to original objects
- Deprecated _PyTrackObjectFactory for generating objects
- Added objects_from_dict and objects_from_array convenience functions
- Reformatted code with black/flake8
- Added .github/workflow
- Added convenience function to convert Pandas or numpy arrays to PyTrackObject
- Added __repr__ for object and tracklet representations in jupyter notebooks
- Deprecated import_HDF in favor of the HDF5FileHandler
- Bug fix for CSV importer
- Added tracker.to_napari() for integration with napari track layer
- Added change to logging verbosity when instantiating tracker
- Added radial clipping option to linker
- Exposed solver options to tracker interface
- Added Windows support
- CSV object importer
- CSV track exporter
- Removed JSON export
- Added children to tracks when loading from hdf
- Added checks to hdf handler to make sure properties exist when loading/saving
- Added is_leaf, start and stop properties to tracklet class
- Added approximate Bayesian updates for use with very large datasets/numbers of objects. Reduces compute time at expense of completeness.
- Added generational depth to the tree output
- Simplify IO/cleaned up HDF handler
- Added 'lazy' termination and initialization hypotheses to aid diagnostics
- Store library version number in HDF files
- Fixed error where last frame of data not added to tracks
- Removed python lineage tree generation (now performed in C++ lib)
- Remove JSON track loader/exporter
- Root nodes now how correct root and parent ID set by track manager
- Makefile infers version number from VERSION.txt during build from source
- Added _build_track_from_dict to JSON loader
- Fixed error writing filtering to HDF
- PyPi release
- Added property filtering to export_delegator, and HDF exporter
- Sanity check to ensure shared library has same version as python wrapper
- Filtering options stored in HDF file now
- Preparations for pip registration
- Store library version in shared lib
- Preparations for pip package
- Improved HDF loader to allow recovery of tracks and trees into native format
- Fixed JSON loader
- Added example tracking data to the repository
- Bug fix to ObjectFactory
- Provided better sample configuration file
- Added a generic filtering option when retrieving objects from HDF files
- Fixes some small bugs and updated documentation
- Added new states {NULL, DUMMY} to PyTrackObject
- Changed default dummy insertion behavior, now given a DUMMY state (removed)
- Added VERSION.txt for quick update of version numbers
- Small update to Python packaging
- Cleaned repo structure, moved load_config to utils
- Fixed bug with default class labels in ObjectFactory
- Added an LBEP exporter for future integration with Napari?
- Unified data export methods, now use tracker.export()
- Migrated to Python 3.7
- Updated (cleaned) Makefile for easier installation
- Simplified hypothesis generation
- Added extra feedback to user including different hypotheses for initialization and termination of tracks
- Tracks can be appended to HDF input files
- Bug fixes to MATLAB exporter
- Fixed bug with sample config and when returning intermediate output of Kalman filter
- Added lineage tree creation to TrackManager in C++ lib
- Added track lineage tree creation
- Simplified code for conversion to Python 3
- Improved HDF reader for performance
- Added back ability to write tracks to HDF file
- Changed tracklet base type to refer to pointers to original objects
- Better handling of object metadata with new Tracklet class
- Added enumetated States, Fates and Errors to python lib
- Cleaned repo structure for deployment
- Add a git clone of eigen during installation (if required)
- Tested install and compilation scripts on linux
- Changed model loading to user defined directory
- Simplified model configuration format
- Allowed a second user model directory, to supplement core models
- Added install scripts
- Changed default apoptosis hypothesis calculation to reflect relative number of observations
- Removed lineage tree generation (now part of Sequitr)
- Improved JSON export
- Added children to return type
- Uses internal track ID for reference
- Added a split track function, using a rule to split
- Add windows compatible __declspec(dllexport) for .DLL compilation (not tested)
- Added set_volume function to define the imaging volume
- Moved btrack types to seperate lib to help migration to python 3
- Added a fast update option that only evaluates local trajectories fully
- Added get_motion_vector function to motion model to make predictions more model agnostic
- Added the ability to select which hypotheses are generated during optimization
- Added more tracking statistics to logging
- Improved track linking heuristics
- Minor bug fixes to log likelihood calculations
- Changed default logger to work with Sequitr GPU server
- Cleaned up rendering of tracks for Jupyter notebooks
- Added time dimension to 'volume' cropping
- Added fate property to tracks
- Returns dummy objects to HDF5 writer
- Returns parent ID from tracks to enable lineage tree creation
- Hypothesis generation from track objects, integration of new Eigen code
- Hypothesis based track optimisation using GLPK
- Track merging moved to C++ code as part of track manager
- HDF5 is now the default file format, for integration with conv-nets
- Tracker returns references to HDF5 groups
- Started integration of track optimiser code
- Set limits on the volume, such that tracks which are predicted to exit the tracking volume are set to lost automatically.
- Enabled frame range in tracking to limit the range of data used
- Fast plotting of tracks
- Output a tracking statistics structure back to Python
- Track iteration to enable incremental tracking and feedback to user
- Major update. Converted Bayesian update code to use Eigen
- Added z-dimension to tracking