Here are the instructions to use the code.
This matlab code is based on the CVPR2012 paper 
Jimei Yang and Ming-Hsuan, "Top-Down Visual Saliency via Joint CRF and Dictionary Learning".
It includes the redistributions of 
1) Shai Bagon'd graphcut code: http://www.wisdom.weizmann.ac.il/~bagon/matlab.html,
2) Liblinear code: http://www.csie.ntu.edu.tw/~cjlin/liblinear/,
3) Jianchao Yang's sparse coding code: http://www.ifp.illinois.edu/~jyang29/ScSPM.htm,
4) Ce Liu's dense SIFT code: http://people.csail.mit.edu/celiu/SIFTflow/,
5) Kevin Murphy's Belief Propagation code: http://www.cs.ubc.ca/~murphyk/Software/CRF/crf2D_usage.html.

This code has been tested on Windows 7 and Ubuntu 11.04 machines.

1. run demo.mat to produce the saliency maps 
   with pretrained bike, car and person saliency models.
2. run run_graz02_slvcrf_bike.m,run_graz02_slvcrf_car.m or run_graz02_slvcrf_person.m 
   to re-train saliency models on the Graz-02 datasets.
   (you need download the graz02 datasets in: http://www.emt.tugraz.at/~pinz/data/GRAZ_02/).

For more information or any questions, please contact Jimei Yang at jyang44@ucmerced.edu.