Lucas Maystre

GIFGIF Data

Page last edited on 2017-07-07

GIFGIF is a project from the MIT Media Lab which aims at understanding the emotional content of animated GIF images. Visitors on the web site are faced with two images and a question of the form: Which image better expresses happiness (pride / fear / …)? In total, 17 different emotions are covered.

Example image from the dataset.

A dataset of pairwise comparison outcomes was graciously provided by the GIFGIF team. It consists of 2.7+ million pairwise comparisons over 6170 animated GIFS, and consists of all the data until January 21st, 2015.

You might also be interested in this Jupyter notebook which illustrates how to read and process the CSV file. Finally, note that this dataset was used in the following paper:

ICML 2017 Lucas Maystre, Matthias Grossglauser
Just Sort It! A Simple and Effective Approach to Active Preference Learning