constantinpape / neurometrics

Bundling different metrics for evaluation of Neurosegmentation

NOTE

This functionality was migrated to nifty

NeuroMetrics

This repo bundles different metrics for evaluating segmentations (especially for neuron segmentation). Implementation in C++ with python bindings.

Implemented:

• randIndex (implementation taken from Bjoern Andres)
• variationOfInformation (implementation taken from Bjoern Andres)
• randFScore (implementation adapted from https://github.com/fiji/Trainable_Segmentation/blob/master/src/main/java/trainableSegmentation/metrics/RandError.java)

Installation

• run cmake in src
• run make in bld
• set pythonpath to bld/python

TODO's

• proper installers (make install, conda)
• Tests
• Documentation
• more efficient: use same contingency matrix for calculating all measures
• Add randPrecision, randRecall and measures derived from VI
• change pybindings to pybind11

neurometrics's Issues

fast weighted metrics

Should be easy as pie to integrate those...mhhhh pieeee

https://github.com/DerThorsten/inferno/blob/master/include/inferno/learning/loss_functions/variation_of_information.hxx
https://github.com/DerThorsten/inferno/blob/master/include/inferno/learning/loss_functions/rand_index.hxx

Buggy VI

For some inputs the VI measures produce non-sensical values.