Implicit Augmentation from Distributional Symmetry in Turbulence Super-Resolution

Requirements

To install requirements:

pip install -r requirements.txt

To get the dataset:

curl -L -o numpy_middle.zip "https://zenodo.org/record/16997999/files/numpy_middle.zip?download=1"
unzip -o numpy_middle.zip -d dataset 
curl -L -o numpy_nearwall.zip "https://zenodo.org/record/16997999/files/numpy_nearwall.zip?download=1"
unzip -o numpy_near_wall.zip -d dataset

To train the model..

For $T=1$ training samples with 1 box at the boundary:

python3 src/train.py configs/boundary_single_timestep_1_box.yaml

For $T=1$ training samples with 1 box at the boundary with explicit rotational augmentation:

python3 src/train.py configs/boundary_single_timestep_1_box_augmented.yaml

For $T=1$ training samples with 3 boxes at the boundary:

python3 src/train.py configs/boundary_single_timestep_3_boxes.yaml

For $T=1$ training samples with 1 box at the midplane:

python3 src/train.py configs/midplane_single_timestep_1_box.yaml

End-to-End examle pipeline:

./main.sh