Just Do It

git clone https://github.com/wangguan1995/3D_ESRGAN_Turbulence.git
cd 3D_ESRGAN_Turbulence
checkout master

wget https://dataset.bj.bcebos.com/PaddleScience/cylinder3D/3D_ESRGAN/data.zip
unzip data.zip

pip install -r requirements.txt     # install libs

python nor_fluc3d.py                 # normalize data
python ESRGAN_3D.py                  # train and plot png

Paper

2022 Three-dimensional ESRGAN for super-resolution reconstruction of turbulent flows with tricubic interpolationbased transfer learning

Code from:

https://fluids.pusan.ac.kr/ fluids/65416/subview.do

Colorful pictures

Overview

3D-ESRGAN is developed to reconstruct 3D super-resolution channel flow from low-resolution data.

Data (Very Poor)

100 snapshots (channel flow at Rer =180) and its low-resolution data are provided for tutorial.  Before the input, the data should be normalized using code nor_fluc3d.py to get normalized data.

Training

Use 3d-esrgan,py to train the deep learning model. It will save architecture and weights files automatically.