PLAS is an algorithm from Self-Organizing Gaussians. It can assign a set of high dimensional vectors onto a 2D grid, and make the result as smooth as possible. That is, for every pixel on the grid, its adjacent pixels have very similar values to itself. For example, if you apply this to a uniform rgb noise image, a 3d uniform distribution is forced to projected onto a 2d plane, it will create an emergement fractal look.

This is a reimplementation of this algorithm. I try to implement it as a single compact function, so that people can understand the algorithm by simply reading the code. Some implementation details may differ from the original one. This leads to some limilations, like it can only process images with widths and heights that can be divided by 2. Neverthless, the result of the algorithm, especially the fractal appearance is very similar to the ones generated by the original algorithm.