This work used deep image priors for CFD flow inpainting.
| Original | Masked | Reconstructed | Energy spectra comparison |
|---|---|---|---|
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
Note: each u, v, w velocity is encoded as a RGB channel for these images.
An illustrative Jupyter notebook, demo.ipynb, is provided in this repository. The quickest way to setup the Python environment is to use pipenv install and then pipenv shell.
Reproducing the turbulence results from the paper entails the following steps:
-
Generating the images from raw turbulence data:
$ python gen_turbulence_images.py -r 64 -
Generating the image masks:
$ python gen_turbulence_masks.py -r 64 -
Running the inpainting script:
$ python inpainting.py --use_gpu -r 64 -i 2001 -
Running the interpolator script with standard GP:
$ mpirun -np 4 python interpolator.py -k gp -r turbulence/results -
Post-processing the results:
$ mpirun -np 4 python analysis_turbulence.py -p -
Generating the plots:
$ python plotter_turbulence.py
For the flow around a cylinder:
-
Generating the images from raw flow data:
$ python gen_cylinder_images.py -r 512 -
Running the inpainting script:
$ python inpainting.py --flow cylinder --use_gpu -r 512 -i 2001 -
Running the interpolator script with standard GP:
$ mpirun -np 4 python interpolator.py -k gp -r cylinder/results -
Post-processing the results:
$ mpirun -np 4 python analysis_cylinder.py -p
@article{HenrydeFrahan19,
author = {Henry de Frahan, M. T. and Grout, R.},
title = {Data recovery in computational fluid dynamics through deep image priors},
journal = {arXiv:1901.11113},
year = {2019}
}
This is work relies on code published and presented here:
@article{UlyanovVL17,
author = {Ulyanov, Dmitry and Vedaldi, Andrea and Lempitsky, Victor},
title = {Deep Image Prior},
journal = {arXiv:1711.10925},
year = {2017}
}
The code that is used here from the deep image prior work can be found in the folder deep_image_prior.







