Skip to content
Branch: master
Go to file


Failed to load latest commit information.

Salient Object Detection - Evaluation Script

Matlab script to evaluate salient object detection algorithms on Achanta's dataset.

If you use any of this work in scientific research or as part of a larger software system, you are kindly requested to cite the use in any related publications or technical documentation. The work is based upon:

[1] Boris Schauerte, Rainer Stiefelhagen, "How the Distribution of Salient
    Objects in Images Influences Salient Object Detection", In Proceedings
    of the 20th International Conference on Image Processing (ICIP), 2013.


The code should be easy to use. You only need to specify the pathes to the dataset, which you can do by setting/changing the variables imagepath and maskpath, and the function handle saliency_func to the saliency algorithm that you want to evaluate.

The MEX-.cpp file analyse_recall_precision_mex.cpp should compile automatically, if neeeded. Of course, you need to have Matlab's "mex" command properly configured.

You are able to specify some further options such as, e.g., a maximum salient object size. Just have a look in the code ...

Most of the evaluation measures that you see at the end are described in the corresponding ICIP paper [1]. You will notice that the script returns multiple results, which is due to the fact that the results depend on the point where you calculate the mean/average. This was something I wanted to keep in the code, because I believe that this may be one of the reasons why the results reported by some authors differ.


The code has two "weak" dependencies. First, the initial example uses my Spectral Visual Saliency Toolbox and, second, you can use Steve Hoelzer's progress bar to visualize the evaluation progress.

If you want to evaluate quaternion-based saliency detection algorithms, then you have to include the QTFM library as well. However, the Spectral Visual Saliency Toolbox should be able to automatically download and install a patched QTFM version for use with QDCT, PQFT, EigenPQFT, etc.

There are three simple ways to handle the dependencies:

  1. Simply run the install_dependencies script, which tries to install the Spectral Visual Saliency Toolbox and the progressbar package to "libs". The "libs" subfolder will be created, if it does not exist. NOTE: The unzip command can be problematic under Mac OS X, in which you should best extract troublesome zip package(s) using another archive utility.
  2. Simply add the library locations (i.e. where you downloaded/unpacked/installed them) to Matlab's search path(es) before you execute the script, see Matlab addpath.
  3. By default the script expects to find the Spectral Visual Saliency Toolbox under "../saliency" and the libraries under "../libs". However, you can of course change these default pathes to suit your setup.


Then you should also check out these two code fragments that are part of [1]



Boris Schauerte


Part of this work was/is supported by the German Research Foundation (DFG) within the Collaborative Research Program SFB 588 "Humanoid Robots" and the Quaero Programme, funded by OSEO, French State agency for innovation.


Salient Object Detection Evaluation



No releases published
You can’t perform that action at this time.