PyTorch implementation of
A Physics-informed Diffusion Model for High-fidelity Flow Field Reconstruction
(Links to paper: Journal of Computational Physics | arXiv)
Denoising Diffusion Probablistic Models (DDPM) are a strong tool for data super-resolution and reconstruction. Unlike many other deep learning models which require a pair of low-res and high-res data for model training, DDPM is trained only on the high-res data. This feature is especially beneficial to reconstructing high-fidelity CFD data from low-fidelity reference, as it allows the model to be more independent of the low-res data distributions and subsequently more adaptive to various data patterns in different reconstruction tasks.
Datasets used for model training and sampling can be downloaded via the following links.
-
High resolution data (ground truth for the super-resolution task) (link)
-
Low resolution data measured from random grid locations (input data for the super-resolution task) (link)
Download the high res and low res data and save the data files to the subdirectory modulus/examples/generative/diffusion/Kolmogorov_2D_data/.
- Note: The directory from which the downloaded dataset files are loaded is specified in the configuration yaml files at
modulus/examples/generative/diffusion/conf/. In the case when the default relative file location in a yaml file cannot be correctly recognized, please replace the relative location with the absolute location. For example, in the configuration filemodulus/examples/generative/diffusion/conf/config_dfsr_train.yaml, Line 24, the value of the key 'data' can be changed to an absolute file directory of the dataset file, e.g.,/<directory of modulus>/examples/generative/diffusion/Kolmogorov_2D_data/kf_2d_re1000_256_40seed.npy
Step 1 - Model Training
In directory modulus/examples/generative/diffusion/, run:
(without physics-informed conditioning)
python train.py --config-name=config_dfsr_train
or
(with physics-informed conditioning)
python train.py --config-name=config_dfsr_cond_train
Step 2 - Super-resolution
In directory modulus/examples/generative/diffusion/, run:
(without physics-informed conditioning)
python train.py --config-name=config_dfsr_generate
or
(with physics-informed conditioning)
python train.py --config-name=config_dfsr_cond_generate
This implementation is based on / inspired by:
- https://github.com/ermongroup/SDEdit (SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations)
- https://github.com/ermongroup/ddim (Denoising Diffusion Implicit Models)