Global General

CODE: GitHub
DATATYPE: image clusters
SOURCE: Instagram
PLACE: Bangkok, Berlin, Moscow, Sao Paulo, Tokyo
YEARS: 2014-2015
SIZE: 50 clusters
CLUSTERS: k-means
SUBSET: "general"
VIZTYPE: histogram
FEATURES: mean hue, brightness
FEAT SRC: scikit-image
PROJECT: Redi, Crockett, and Manovich. ACMMM '16
MORE INFO: Software Studies Blog

This is an "image histogram", which gives the distribution of a variable and uses images as plot elements. Because images carry information beyond what is measured on the histogram axis, additional vertical sorting is actually useful and important. Here, the plot elements are not single images but clusters of images. This is an important bit of wisdom about direct visualizations like these: the raw data are always images, but there are no restrictions on which transformations of these are allowable. You can use images, clusters of images, slices of images, clusters of slices, annotated images, etc. (we've used all of these in different ways). The plot here is an early attempt at cluster analysis, superseded in some ways by the growing entourage plot, although here we downplay the differences in cluster size and perhaps get a clearer sense of their visual profile. We noticed, for example, that some of the clusters needed refinement, and so we changed our clustering method and got better results.