Code for generating the dataset and evaluation metrics discussed in Confluent Vessel Trees with Accurate Bifurcations which will appear in CVPR2021. The implementation of the data generation was forked from https://github.com/midas-journal/midas-journal-794 and the original project description is at http://www.insight-journal.org/browse/publication/794.
The script is tested under python 3.5.2. To install the dependencies, run
pip install -r requirements.txt
To obtain the ROC and angular error curves, use the exmaple script eval.sh. The input will be the reconstructed tree in H5 format (Our reconstructed trees (corresponding to the green curve below) are provided here.) and groundtruth in XML format. The output will be the average measure over the 15 volumes and saved in a file named "***Average.csv". This file contains the following variables.
To generate the "On whole tree" ROC curve,
X-axis: 1 - PercentageOfPointsCloserThanRadius
Y-axis: PercentageOfPointsCloserThanRadiusOrig
To generate the "On branching points" ROC curve,
X-axis: 1 - PercentageOfBifurPointsCloserThanRadius
Y-axis: PercentageOfBifurPointsCloserThanRadiusOrig
To generate the angular error plot,
X-axis: ThresholdValue
Y-axis: averageAngleAtBifur