Welcome to evalsaliency, a Matlab toolbox for evaluating salient object detection algorithms. This toolbox has been utilized to achieve experimental result presented in the following paper:
##Xi Li, Yao Li, Chunhua Shen, Anthony Dick, Anton van den Hengel. Contextual Hypergraph Modeling for Salient Object Detection, ICCV 2013. PDF | Project page.
Please cite this paper if you use this toolbox in your research.
To get started, please ensure that MATLAB has been installed and then download the toolbox from Github. The toolbox has been tested on Linux.
##Evaluation criteria included
The current version of the evalsaliency toolbox contains five evaluation criteria used in salient object detection algorithms, including precision-recall curve (PR curve), receiver operating characteristic curve (ROC curve), F-measure, mean absolute error (MAE) and VOC overlap score.
##Folders and files
The "data" folder contains mean-shift segmentation of each image in a dataset. We have provided precomputed segmentation of two benchmark datasets, MSRA1000(ASD) and BSD300(SOD).
The "result" folder contains five subfolders, corresponding to result of five evaluation criteria.
"matlabPyrTools", "feature_util" and "edison_matlab_interface" are thee folders contain files to compute mean-shift segmentation.
"eval*.m" are Matlab scripts for computing the five evaluation criteria. "draw*.m" are Matlab scripts for displaying result.
##Computing PR curve, ROC curve and MAE
You need to change four variables in the corresponding "eval*.m" , including "method", "dataset", "resultpath" and "truthpath" and then run the script.
##Computing F-measure and overlap score
As both F-measure and overlap score are rely on mean-shift segmentation, you need to ensure segmentations of the dataset are exist in the "data" folder. If exist, you then change the aforementioned four variables before running the script. If not, you need run "ExtractMeanShiftSegmentationMask.m" first to compute mean-shift segmentation.
##Displaying result
After computing "eval*.m", you can run "draw*.m" to see figures. Keep in mind that you also need to change the four variables in the "draw*.m".
##Questions and feedback
If you have any questions or feedback, please email to [email protected]