St. Petersburg Meals
| CODE: |
GitHub |
| DATATYPE: |
images |
| SOURCE: |
Instagram |
| PLACE: |
St. Petersburg |
| YEARS: |
2014-2015 |
| SIZE: |
2,961 |
| SUBSET: |
"meal" |
| VIZTYPE: |
growing entourage gridded |
| FEATURES: |
tag space |
| FEAT SRC: |
Google Cloud Vision |
| ORIG FEAT DIMS: |
697 |
| PROJECTION: |
t-SNE |
| PROJECT: |
St. Petersburg Instagram |
| ANNOTATION |
cluster tags |
This is a "growing entourage" plot of Instagram images in the category "meals". Such plots take images located in a high-dimensional similarity space, cluster them by some method, and project the clusters to a 2D canvas. Here, the images are grouped by machine tags from Google's Cloud Vision API. The centroids of each cluster are then projected to 2D by t-SNE, and the centroids then take turns adding members to their "entourages", in order of visual similarity to the centroid. New members occupy the open square nearest their centroids. Territorial conflicts are settled by this principle. Centroid locations are annotated with the tag that defines the cluster. This plot is somewhat different than a classic growing entourage, because images can officially belong to multiple clusters. As each entourage has its turn to add a member, it chooses whichever image, among those with the relevant tag, that is most similar to its centroid. This results in more similarly-sized clusters than is usual for a growing entourage.