pubfig_attribute_rationale_annotations.txt is a CSV file containing the data from our experiment on the Public Figures dataset [1], which can be downloaded from:
	http://www.cs.columbia.edu/CAVE/databases/pubfig/download/pubfig_attributes.txt

The first line contains the column names.

The first two columns are the ID and public figure's name, referring to those in Kumar et al.'s data.

is_hot column is a 1 or 0 depending on whether we (the authors) decided the celebrity was "hot" or "not" (respectively).  For any given person, this will always be the same.

is_male column is a 1 or 0 depending on whether the person is a male or female (respectively).

annotator_id column is a unique numerical ID corresponding to the human annotator that performed the annotation.

annotation_id column is simply a unique ID for the row.

annotation_start_time and annotation_end_time are when the user started and finished working on the annotation, in seconds since the Unix epoch.  Note that annotation_end_time - annotation_start_time may not be an accurate reflection of the time the annotator spent working on the annotation, as it is simply based on when they requested the annotation page and when they submitted their work.

The remaining columns are the actual rationale for each attribute: 0 if the attribute was not chosen as a rationale, 1 if the attribute was chosen as a rationale.


[1] N. Kumar, A. C. Berg, P. N. Belhumeur, and S. K. Nayar. Attribute and Simile Classifier for Face Verification. In ICCV, 2009.