URecon is a simple UNET architecture implemented in TensorFlow that can be used to reconstruct the initial conditions of N-body simulations from late time (e.g. z=0) density fields. The weights trained on Quijote fiducial cosmology simulations, used for the results presented in 2305.07018 are provided.
Requirements:
-
numpy/scipy
-
tensorflow
-
tensorflow_addons (for instance normalization)
-
tqdm
-
pylians3 (only for measuring power spectrum of density fields)