Denoising Diffusion Probabilistic Model (DDPM) in Pytorch to generate CAMELS astrophysical maps.
Download images from the CAMELS Multifield dataset.
In this example we make use of 15k maps of the total mass field at Maps_Mtot_SIMBA_LH_z=0.00.npy
Two different implementations are included:
-
camels_diffusion_model_from_scratch.ipynb: an implementation from scratch adapted from the tutorial Diffusion models from scratch in PyTorch by DeepFindr. -
ddpm_camels.ipynb: using a more updated implementation of diffusion models from Denoising Diffusion Pytorch. Convert CAMELS data to a dataset of images usingget_camels_maps.ipynb.
To validate the accuracy of the generated images, one can compute several summary statistics such as the power spectrum or the probability density function, in summary_stats.ipynb.
