Скачать или смотреть K-Means Color Segmentation of Starship Launch

This is a demonstration of K-means clustering. K-means is an algorithm that partitions data into k groups by iteratively minimizing the cumulative distance between points and their assigned centroids. Each point is assigned to the nearest centroid, and the centroids are updated to the mean of their assigned points. After several iterations, the centroids stabilize and the data is segmented into clusters.

In this demonstration, 20 centroids segment about 2 million pixels per frame of a SpaceX Starship launch video. The pixel values are projected into RGB color space (shown below the video). Unlike typical usage, where k-means runs to convergence, here the algorithm advances only one iteration per frame. This reveals how the centroids move dynamically as the algorithm progresses and as the underlying data changes.

There are many possible extensions. For example, points could be made semi-transparent to better visualize density in color space. More artistically, the centroids could be given momentum and move under elastic forces rather than snapping directly to the mean. Perhaps even the pixels themselves could evolve under physical transport equations, or even be made to appear to orbit the centroids. Food for thought.

#kmeans #segmentation #datascience #starship