Working with large datasets of images one common task is to order or cluster them by similarity. Computer science offers a plethora of algorithms and tools to help with that purpose. A very popular and useful algorithm is t-sne which clusters similar items into meaningful clouds. The problem though is that when you apply this to images you get a lot of overlapping items that often prevent you from inspecting a single image.

I tried to solve this problem by finding an algorithm that transform these cluster clouds into a regular grid, but without destroying the neighborhood relations that emerged from the clustering. Raster Fairy is the result of my research.