Neural networks weights related to "Super-resolution of turbulence with dynamics in the loss"

A set of routines to train neural networks to perform super-resolution, without necessarily requiring high-resolution data, along with network weights used to generate figures in the publication "Super-resolution of turbulence with dynamics in the loss", (J. Page, Journal of Fluid Mechanics, accepted 2024). Implementation wraps around the spectral version of JAX-CFD (https://github.com/google/jax-cfd). Neural networks are written in Keras using the JAX backend. This dataset consists of a series of python scripts and .h5 files of network weights, organised into subdirectories: code/ # scripts used to train and analyse the networks (python, requires installation of keras + JAX backend, JAX-CFD) paper_weights/ # weights for the neural networks as documented in the paper. # subdirectories here point to different Reynolds numbers (100, 1000) with weights at M = 16 and M = 32 coarse-graining # an additional subdirectory includes networks trained with noisy data Neural network weights are saved as .h5 files. An example script is included ("load_weights.py") to illustrate how to load into a model.