1 Introduction
This document describes the mode of operation of the evaluation code supplied with
the Weizmann segmentation evaluation database.
2 Basic setup
The database contains one folder for each image, where each of which contains the
human segmentations and the source image in separate folders. To use the functions
supplied one must create a folder in each image folder that contains the output of the
segmentations one whishes to evaluate.
Figure 1: Segmentation evaluation folders tree
The output of the evaluated method should be saved in image format where each pixel
value represents the class number to whom it belongs. It is recommended to use either
8-bit or 16-bit PGM format.
Figure 2: Example of the evaluation output format
Note that if the results folder contains more then one output files the evaluation
methods selects the one with the highest F-measure score.
3 Evaluations functions
Two Matlab evaluation methods are supplied with the database. The methods
implements the two tests described in our CVPR paper.