METHOD FOR MEASURING THE BOUNDARY PERFORMANCE OF A SEMANTIC SEGMENTATION NETWORK

摘要

The invention relates to a method for evaluating the boundary performance of a semantic segmentation network in a convolutional neural network.;Method for measuring the boundary performance of a semantic segmentation network comprising the steps:determining (S1) an original boundary mask (Bo) defining a boundary between different classes of an image;determining (S2) a thickened boundary mask (Bt) dependent on the original boundary mask (Bo) using morphological operations;determining (S3) a true positive rate, a false positive rate and a false negative rate dependent on the original boundary mask (Bo) and/or the thickened boundary mask (Bt); anddetermining (S4) a final boundary metric, indicating the boundary performance, dependent on the determined true positive rate, the determined false positive rate and the determined false negative rate.