Pascal Objects Eye Tracking (POET) V 1.1
========================================

Dim P. Papadopoulos, Alasdair D. F. Clarke, Frank Keller and Vittorio Ferrari
dim.papadopoulos@ed.ac.uk
 

Introduction
~~~~~~~~~~~~
 
Welcome to this release of the Pascal Objects Eye Tracking (POET) dataset!
We collected eye tracking data for the complete trainval set of ten objects classes 
(cat, dog, bicycle, motorbike, boat, aeroplane, horse, cow, sofa, diningtable)
from Pascal VOC 2012 [2] (6,270 images in total). Each image is annotated with the eye movement 
record of five participants, whose task was to identify which object class was present in the image 
(see [1] for more details).

This dataset was first published in [1]. Please cite it if you use this dataset.


Contents
~~~~~~~~

Let <dir_root> be the directory where this package was uncompressed.

This package contains:

 - 6270 Pascal VOC 2012 trainval images from [2] in <dir_root>/PascalImages

 - 10 mat files (one for each class) with eye tracking data in <dir_root>/etData

 - 20 txt files (two for each class) with the splits of images into R_{bb+fix} and R_{fix} sets in <dir_root>/split (see [1] for more details about the sets)

 - matlab code to visualize fixations and fixation heatmaps in <dir_root>/code
 

Quick start
~~~~~~~~~~~
 
You can follow the next steps to check that everything is working properly:
 
1) start matlab 
 
2) navigate to <dir_root> (e.g. by using the 'cd' command)
  
3) execute command: demo_drawFixations
   This demo visualizes the eye tracking data on images. 
   Press any key to continue to the next image while the figure window is active.

4) execute command: demo_drawFixHeatMap
   This demo visualizes the heatmap of the eye tracking data on images. 
   Press any key to continue to the next image while the figure window is active. 

5) if all above points went well, this package is working perfectly.

 
Description of the etData files
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
Each mat file in <dir_root>/etData contains one structure array with the eye tracking data for this class:

    filename - filename of an image in Pascal format
    dimensions - a 1x2 double array with image dimensions, [height width]
    gtbb - an Nx4 double array with the coordinates of the N ground-truth bounding-boxes, [x1 y1 x2 y2]
    rt - an 1x5 double array with the response times in seconds of the 5 different viewers
    response - a 1x5 logical array with the correct button presses.
    imScalingFactor - the scaling factor used to diplay the image in the monitor
    displayOffset - the coordinates of the window where the image was diplayed in the monitor with resolution 1680x1050
    
    fixations - a 1x5 structure array with all information about the eye tracking data of 5 viewers
        scrnCoord - screen coordinates of the eye tracking data 
        imgCoord - image coordinates of the eye tracking data

        fixR - fixations of the right eye
        fixL - fixations of the left eye

        time - an Mx2 array with the initial and final time of the M fixation in ms
        pos - an Mx2 array with the x and y coordinates of the fixation positions.
        

Description of the split files
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Each txt file (class_set.txt) in <dir_root>/split contains a list of image filenames (in Pascal format) that this set (Rfix or Rbbfix) of this class contains.
The fixation-to-bounding-box model in [1] was trained on the images in the R_{bb+fix} set. After training, this model was used to predict bounding-boxes for the images in the R_{fix} set.
 
Matlab code
~~~~~~~~~~~
 
The following Matlab functions are provided in <dir_root>/code:
- drawFixationOnImg: draws the eye tracking on an input image
- drawFixationHeatMap: draws a heatmap of the eye tracking data on an input image
- getFixationPositions: returns all eye tracking positions for an input image.
 

Support
~~~~~~~
 
For any query/suggestion/complaint or simply to say you like/use the dataset and software just drop us an email
 
dim.papadopoulos@ed.ac.uk
keller@inf.ed.ac.uk
vferrari@staffmail.ed.ac.uk
 
 
References
~~~~~~~~~~

[1] Training Object Class Detectors from Eye Tracking Data
D. P. Papadopoulos, A. D. F. Clarke, F. Keller and V. Ferrari 
European Conference in Computer Vision (ECCV), September 2014.

[2] The PASCAL Visual Object Classes Challenge 2012 (VOC2012) Results
M. Everingham, L. Van Gool, C. K. I. Williams, J. Winn and A. Zisserman
http://www.pascal-network.org/challenges/VOC/voc2012/workshop/index.html


Version History
~~~~~~~~~~~~~~~

Version 1.1 (March 2015)
-----------
- Files with the split of images into R_{bb+fix} and R_{fix} sets were added

 
Version 1.0 (September 2014)
-----------
- initial release