This is a simple TensorFlow 2 implementation of Fourier Feature Networks, as described in this paper. Also see their project page.
The code here generates random Fourier features to use in a compositional pattern producing network (this is the 2D image regression task described in the paper.)
The config.yaml
file controls hyperparameters, which are kept at their
defaults. The data is 6 images taken from places365
.
The top row is a standard compositional pattern producing network, and the second row is the same network using random Fourier features.