This folder contains annotation data for the MSRC version 1 dataset collected 
from anonymous users using the Amazon Mechanical Turk interface. 

For each image in the MSRC dataset there is a corresponding mat file with the 
same name (eg: 1_10_s.mat) that contains object oultines (polygons) and 
time taken to provide annotations by a list of users. There are about 70 users
in total (after manually pruning incorrect segmentations). Each image has 
annotation data from about 20-30 users. Note that not all users worked on all 
the images. 

Description of .mat files (1_10_s.mat)
-----------------------------------------------------------------------------
Each .mat file contains the following variables

1. annotation_times is a vector recording the time taken by a list of users to
   completely outline and label the objects in the image

2. annotation_data is an array of structs (one for each user who worked on the 
   image) containing the following variables

   imgname - name of the image
   userid - the user's id from 1 to 70
   names - is a cell array containing the names of objects outlined
   	   by the user
   times - is a vector recording the time taken to outline each object
   total_time - the total time taken on the image (sum of the 
   	      	times array)
   polygons - cell array of Nx3 matrix where N is the number of points clicked
   	      on the object	
   	      The three columns record
	      x - x coordinate of point in the image
	      y - y coordinate
	      t = time in milliseconds when point was clicked
	      ie. polygons{1}(end,3) - polygons{1}(1,3) = time taken to outline object 1.
   
The folder also contains a file called user_data.mat which records the images worked 
on by a particular user. The file contains a single array of struct (one for each user 
indexed by userid defined above) containing fields imgname, total_time.


Please refer the paper "What's It Going to Cost You?: Predicting Effort vs. 
Informativeness for Multi-Label Image Annotations", S. Vijayanarasimhan and K. Grauman,
CVPR 2009 for more details.