Accelerating the estimation of effective rock properties under partially saturated conditions through saturation-conditioned U-Net–based initialization

This repository contains datasets for constructing a saturation-conditioned U-Net to distribute the gas phase in porous media. The data_pool folder contains the raw 3D datasets, while the raw_data_k1 and raw_data_s1 folders contain the training, validation, and prediction datasets used for machine-learning model construction. The script folder contains Python scripts used to build the machine-learning model, perform hyperparameter optimization, and generate predictions. The template folder provides a Python script template for machine-learning model construction. For detailed Python codes, please check our Github repository: https://github.com/FZJ-RT/cac.git