Reproducible material for ** Post-FWI Injection of Learned Priors Using a Flow Matching Model **

Project structure

This repository is organized as follows:

• 📁 package: Core Python modules implementing the U-Net architecture and Flow Matching framework.

• 📁 asset: Project-related visual assets, including logos and illustrative figures.

• 📁 data: Viking dataset, containing a FWI result.

• 📁 notebooks: Jupyter notebooks for reproducing the otway experiment presented in this work.

Notebooks

The following notebooks are provided:

• 📙 Example_viking.ipynb: notebook performing prior injection test on Viking Dataset;

Trained Model

Download the pretrained model from the provided link: Google Drive. Place the downloaded file in the notebooks/checkpoints/ directory.

Getting Started 👾 🤖

To ensure reproducibility, we recommend creating the environment using the provided environment.yml file.

Run the following command:

./install_env.sh

The installation may take a few minutes. After installation, run.

Activate the environment using:

conda activate postFWI

You can start testing by running the notebook in the ./notebooks/ directory:

Disclaimer: All experiments have been carried on a Intel(R) Xeon(R) Gold 6230R CPU @ 2.10GHz equipped with a single NVIDIA RTX A6000 GPU. Different environment configurations may be required for different combinations of workstation and GPU.