Dan Zarrella gathered a dataset of 1.5 million Instagram photos from over half a million different users. And analyzed the images including their tags, captions as well as the actual visual content of the photographs to identify the characteristics that make images work (or not work).
