panoptica 0.5.12

Panoptic Quality (PQ) computation for binary masks.

Panoptica

Computing instance-wise segmentation quality metrics for 2D and 3D semantic- and instance segmentation maps.

Features

The package provides 3 core modules:

1. Instance Approximator: instance approximation algorithms in panoptic segmentation evaluation. Available now: connected components algorithm.
2. Instance Matcher: instance matching algorithm in panoptic segmentation evaluation, to align and compare predicted instances with reference instances.
3. Instance Evaluator: Evaluation of panoptic segmentation performance by evaluating matched instance pairs and calculating various metrics like true positives, Dice score, IoU, and ASSD for each instance.

Installation

The current release requires python 3.10. To install it, you can simply run:

pip install panoptica

Use Cases

All use cases have tutorials showcasing the usage that can be found at BrainLesion/tutorials/panoptica.

Semantic Segmentation Input

Although for many biomedical segmentation problems, an instance-wise evaluation is highly relevant and desirable, they are still addressed as semantic segmentation problems due to lack of appropriate instance labels.

Modules [1-3] can be used to obtain panoptic metrics of matched instances based on a semantic segmentation input.

Jupyter Notebook Example

Unmatched Instances Input

It is a common issue that instance segementation outputs have good segmentations with mismatched labels.

For this case modules [2-3] can be utilized to match the instances and report panoptic metrics.

Jupyter Notebook Example

Matched Instances Input

Ideally the input data already provides matched instances.

In this case module 3 can be used to directly report panoptic metrics without requiring any internal preprocessing.

Jupyter Notebook Example

Citation

If you have used panoptica in your research, please cite us!

The citation can be exported from: TODO