Data Format    
The data for the evaluation is available as a tar file called eval_data.tar.gz. It untars into a directory called eval_data which in turn consists of three  sub directories - eval_images, eval_segs and eval_labels. These sub directories can also be individually obtained from the links provided below (Note that they are in the form of tar files that can be extracted into individual directories). Below, is a brief description of each of these sub directories and the files they contain illustrated with the help of sample files from each sub directory.

eval_images  -  which has the images that were used for the labeling in JPEG format.  A sample image from this directory can be seen below.
Sample image from the data.

Example image 100039.jpg

eval_segs  - which contains the segmentation data for the above images. These files have the extension .seg.The data is in the form of a text file. Each file has some initial meta information and then a series of numbers as follows:

Sample data from a .seg file

Sample text from file 100039.seg

The meta data gives information on the name of the image, dimensions and number of segments. The row of numbers below this provide information on what pixels belong to what segments. They follow the format

s r c1 c2

s  is the segment id, r is the row and c1, c2 are the start and end column values. Thus, these four numbers give the segment id of all the pixels that lie on row r and between columns c1 and c2. A simple program can be used to read the numbers from this file and produce the region segmentations for the given image. The labeling software has a routine that does this. The segmentations will look as shown below:

Sample image from the data hsowing region segmentations.

Example image 100039.jpeg with region segmentations seen.

eval_labels - which contains the label data. These files have the extension .lab. The label data again has some meta information followed by the number of labels/segments and a list of the labels.

Sample data from a .lab file

Sample label file 100039.lab

The label file is read in by the software and is edited as the labels are changed by the user. These labels are assigned based on a set of labeling rules . Words are used with specific senses and the sense in which a given word is used is indicated in parenthesis by the side of the word. However, no such indication is provided if the word is used in its first sense. For instance, in the set of labels above, we see that  the word sky is used in its first sense whereas branch is used in the second sense.