This program is a python implementation of our CVPR 2013 paper: "Saliency Detection via Graph-Based Manifold Ranking". For c++ implementation, please go check Chuan's github page at https://github.com/yangchuancv/ranking_saliency

You are welcome to download the code and use it in your own research or project. If you publish your work please cite my github URL and our paper:

@inproceedings{yang2013saliency,
title={Saliency detection via graph-based manifold ranking},
author={Yang, Chuan and Zhang, Lihe and Lu, Huchuan, Ruan, Xiang and Yang, Ming-Hsuan},
booktitle={Computer Vision and Pattern Recognition (CVPR), 2013 IEEE Conference on},
pages={3166--3173},
year={2013},
organization={IEEE}
}

The program is tested on both Linux ( debian Jessie) and Mac OS ( El Capitan 10.11 ). If you are using cygwin under windows, it should also be no problem to run the code ( I don't have Windows, so it is just my guess :-) ).

Install

• [xkunglu version 0.0.1] converted to pip installer. After git clone

$ cd mr_saliency $ pip install .

The installer should install all dependencies. With one exception, wxPython ( curr version 4.0.1) fails to build wheel, on Ubuntu 16.0.4 therefore it needs to be installed separately. If you are using this within a conda environment the conda install works:

$ conda install -c anaconda wxpython

For Mac user, you can install wxpython by brew

$ brew install wxpython

If you are using python under virtual environment such as pyenv, you may have problem in using wxpython, get error message like: "This program needs access to the screen. Please run with a Framework build of python, ..". No worry, reinstall your current python under pyenv by

$ env PYTHON_CONFIGURE_OPTS="--enable-framework" pyenv install x.x.x

Dependencies

• Python > 2.7
• wxPython
• opencv > 2.4
• numpy
• scipy
• skimage
• matplotlib

The code is not very sensitive to version of libraries used ( I think, :-)), so just install the dependencies simply by general package managers of your OS (aptitude brew, or pip in a python way).

Usage

Syntax of original command is unchanged

• MR.py There are two classes you can use. One is MR_saliency which is the implementation of manifold ranking saliency.

  $ python
  >>> import MR
  >>> import matplotlib.pyplot as plt
  >>> mr = MR.MR_saliency() # initialization
  >>> sal = mr.saliency(<valid image path>)
  >>> plt.imshow(sal,cmap='gray') # if you want to see result in gray mode
  >>> plt.show()
  

I leave all the parameters open to manipulating, you can set your own parameters during initializing the class. However, you don't have to do it, default values work pretty well, unless you really know what you want to do.¹

Another class is a helper class help to show some calculation results such as superpixels, boundary priors and so on. It is very easy to understand the functionality, go check the source code

• live_demo.py

This is simple live demo showing how to use MR saliency. It is a real time saliency detection application by using webcam²

Bug Report and License

If you find any bug, feel free to report it on github or just throw me an email.

Please Note This program is released under GPL2.0, however, manifold ranking algorithm itself have been patented by OMRON co., Japan, my previous employer, so please don't use it in any commercial product.

Other Resources

• Please also check our saliency detection dataset DUT-OMRON at http://saliencydetection.net/dut-omron
• If you are interested in me ( I guess you are not, :-), go check my homepage at http://ruanxiang.net

Footnotes

1. check paper for details of the parameters ↩

2. the program use device ID 0 as input, you can modify device id to your own device ↩