For the visualization, the utils/process_images.py produces plot_data.json file. In the program, the write_json() function calls get_centroids() which calculates a k-means clustering in the reduced space for the purposes of selecting an image to "summarise" each of the clusters, as "hotspots". However, the clustering is then discarded.

It might, or it might not, be nice to keep the clustering results of k-means and store it for further use.