General:

use the environment.yml with anaconda.

Rerun experiments from the PermeabilityNets paper:

1. Download the data here: https://figshare.com/s/d403ed0cb9816507cb4c and the checkpoints here: https://figshare.com/s/3bce79cf4a47ab0986ae The data is split into a Training, Validation and Testing set. Note that at least 128 GB of RAM and a capable graphics card is required to run the full training! The evaluation should run on 16 GB RAM and a GPU with at least 8 GB of vRAM.

2. Open ModelTrainerScripts.flowfront_to_fiber_STANDALONE.py. This is a version of our training script that is modified to work as a demo script. Only very little setup is needed:

   1. save_path should be a Path() that specifies the output folder. Training and Eval results will be saved there.
   2. data_folder should be a Path() that points to the folder containing the three data splits downloaded in step 1.
   3. model_chkpts should be a Path() that points to the folder containing the four model checkpoints (from step 1.). Do not rename the checkpoints, since it will break the demo script!
   4. mode: choose between Conv2D, Conv3D, Transformer, or ConvLSTM.
   5. eval: set to False if you want to run the full training. For the ConvLSTM and Conv3D model this could take up to a week, depending on your ML-hardware.

3. Run your configuration.

A Translation for the Code:

• ConvLSTM is called FFTFF in Models.sensor_to_fiber_fraction_model.py
• Transformer is called OptimusPrime in Models.flowfront2PermTransformer.py
• Conv2D is called FF2Perm_Baseline in Models.flowfront2PermBaseline.py
• Conv3D is called FF2Perm_3DConv in Models.flowfront2PermBaseline.py