Change log


First public release


Drop compatibility with Python 2.7.

Add replace argument for class.

Mark functions in segmetrics.boundary as private.

Mark segmetrics.boundary.ObjectBasedDistance.obj_mapping as private.

Remove segmetrics.legacy.

Rename segmetrics.metric.Metric to segmetrics.measure.Measure.

Rename segmetrics.boundary.ObjectBasedDistance to segmetrics.boundary.ObjectBasedDistanceMeasure.


Add segmetrics.cli.

Add segmetrics.Measure.accumulative parameter and makes detection-based measures non-accumulative by default.

Remove the segmetrics.Measure.ACCUMULATIVE field.


Change default names of performance measures in

Add segmetrics.measure.Measure.default_name() method.


Add quantile-based implementation of segmetrics.boundary.Hausdorff.


Pass positional and keyword arguments for segmetrics.boundary.DistanceMeasure.object_based through to ObjectBasedDistanceMeasure.


Add --semicolon CLI option.


Add the aggregation keyword argument and member variable of the segmetrics.measure.Measure base class.

Remove the accumulative keyword argument and member variable of the segmetrics.measure.Measure base class.

Remove the FRACTIONAL class variable of the segmetrics.measure.Measure base class.

The method add_measure` of the class now returns the name of the measure.


Change default name of segmetrics.regional.ISBIScore if min_ref_size is not the default.

Change default name of segmetrics.boundary.ObjectBasedDistanceMeasure if skip_fn is not the default.


Rename segmetrics.boundary to segmetrics.contour.

Move segmetrics.detection.COCOmAP to segmetrics.deprecated.COCOmAP.

Minimum supported Python version is now 3.8 due to type linting.

Replace the ObjectBasedDistanceMeasure class by the more general segmetrics.measure.ObjectMeasureAdapter class.

Add .object_based() method for all image-level measures (including region-based measures). The method does not accept positional arguments any more, only keyword arguments.

Remove the argument and attribute mode from the Hausdorff distance.

Add .reversed() and .symmetric() methods for all asymmetric measures (e.g., object-based).

Add geometric-mean aggregation method.

Rename aggregation method obj-mean to object-mean.