Libraries

Learnable Fourier Positional Encoding

PyTorch implementation of learnable Fourier features for spatial positions

Multidimensional coordinates transformed into waves and a learned feature embedding

This package implements Learnable Fourier Features for Multi-Dimensional Spatial Positional Encoding in PyTorch. It transforms multidimensional spatial coordinates into expressive embeddings using Fourier features whose projection is learned with the rest of the model.

The encoding is useful when a neural network needs to reason about positions in more than one dimension while preserving spatial relationships. The library packages the method as a reusable PyTorch module and is available directly from PyPI.