Can denoising diffusion probabilistic models generate realistic astrophysical fields?

This repository contains code for the experiments in Can denoising diffusion probabilistic models generate realistic astrophysical fields?, accepted at the Neurips Machine Learning and Physical Sciences workshop.

We use code blocks and architecture from

• Hugging Face's The Annotated Diffusion model
• Denoising Diffusion Pytorch
• Denoising Diffusion Probabilistic Models

Experiments

Experiments were kept track of using the Weights and Biases framework.
Cold Dark Matter Density Fields

1. For the 64x64 runs, run python main.py config/params64_alt.yaml
2. For the 128x128 run, run python main.py config/params128_blseed.yaml

Images from SFD
3. Run python main.py config/params_dustmm.yaml

An earlier version of this repository contained forked code from Improved Denoising Diffusion Models.