This program is open source under the BSD-3 License. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.

2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.

3. Neither the name of the copyright holder nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

OpenFWI Benchmarks

Pytorch Implementation on OpenFWI 2D datasets

About

This repository officially supports the reproducibility of OpenFWI Benchmarks[1]. The mateirals will evolve with the further development of OpenFWI.

For the time being, it contains the codes for training and testing InversionNet[2] and VelocityGAN[3], and covers 10 datasets in Vel family, Fault family, and Style family.

Prepare Data

First download any dataset from our website and unzip it into your local directory.

Load a pair of velocity map and seismic data

For any dataset in Vel, Fault, Style family, the data is saved as .npy files, each file contains a batch of 500 samples. datai.npy refers to the i-th sample of seismic data. To load data and check:

import numpy as np
# load seismic data
seismic_data = np.load('data1.npy')
print(seismic_data.shape) #(500,5,1000,70)
# load velocity map
velocity_map = np.load('model1.npy')
print(velocity_map.shape) #(500,1,70,70)

Prepare training and testing set

Note that there are many ways of organizing training and testing dataset, as long as it is compatible with the DataLoader module in pytorch. Whichever way you choose, please refer to the following table for the train/test split.

Dataset	Train / test Split	Corresponding .npy files
Vel Family	24k / 6k	data(model)1-48.npy / data(model)49-60.npy
Fault Family	48k / 6k	data(model)1-96.npy / data(model)97-108.npy
Style Family	60k / 7k	data(model)1-120.npy / data(model)121-134.npy

A convenient way of loading the data is to use a .txt file containing the location+filename of all .npy files, parse each line of the .txt file and push to the dataloader. Take flatvel-A as an exmaple, we create flatvel-a-train.txt, organized as the follows, and same for flatvel-a-test.txt.

Dataset_directory/data1.npy
Dataset_directory/data2.npy
...
Dataset_directory/data48.npy

To save time, you can download all the text files from the splitting_files folder and change to your own directory.

Reproduce the OpenFWI Benchmarks

For InversionNet and VelocityGAN, the current version supports training with a single GPU. For UPFWI and InversionNet3D, multiple-GPU is necessary due to the computation cost.

Environment setup

The following packages are required:

• pytorch v1.7.1
• torchvision v0.8.2
• scikit learn
• numpy
• matplotlib (for visualization)

If you other versions of pytorch and torchvision, please make sure they align.

InversionNet

To train from scratch on Flatvel-A dataset with $\ell_1$ loss, run the following codes:

python train.py -ds flatvel-a -n YOUR_DIRECTORY -m InversionNet -g2v 0 --tensorboard -t flatvel_a_train.txt -v flatvel_a_val.txt

-ds specifies the dataset, -n creates the folder containing the saved model other log files, -g2v sets the coefficient of $\ell_2$ loss to be zero, -t and -v assign the training data and test data loading files.

To continue training from a saved checkpoint, run the following codes:

python train.py -ds flatvel-a -n YOUR_DIRECTORY -r CHECKPOINT.PTH -m InversionNet -g2v 0 --tensorboard -t flatvel_a_train.txt -v flatvel_a_val.txt

Please refer to the details of the codes if you would like to change other parameters (learning rate, etc.). These commands suffice to reproduce the OpenFWI benchmarks.

The last step would be testing, where we include the visualization. Also we borrow the implementation of SSIM metric from pytorch-ssim. Please make sure that pytorch-ssim.py and rainbow256.npy are placed together with others.

python test.py -ds flatvel-a -n YOUR_DIRECTORY -m InversionNet -v flatvel_a_val.txt -r CHECKPOINT.PTH --vis -vb 2 -vsa 3

--vis enables the visualization and creates a folder with the figures, you may also change the amount of velocity maps by playing with -vb and -vsa.

VelocityGAN

The code logic of VelocityGAN is almost identical the that of InversionNet.

To train from scratch on Flatvel-A dataset with $\ell_1$ loss, run the following codes:

python gan_train.py -ds flatvel-a -n YOUR_DIRECTORY -m InversionNet -g2v 0 --tensorboard -t flatvel_a_train.txt -v flatvel_a_val.txt

To continue training from a saved checkpoint, run the following codes:

python gan_train.py -ds flatvel-a -n YOUR_DIRECTORY -r CHECKPOINT.PTH -m InversionNet -g2v 0 --tensorboard -t flatvel_a_train.txt -v flatvel_a_val.txt

The command for testing is the same with InversionNet

Future Updates

• We will release the training configuration of Kimberlina-CO2 dataset very soon.
• We will improve the instruction with illustrations and other necessary details
• The codes of UPFWI and InversionNet3D is pending approval, they will be added to this repo once approved.

References

[1] Deng, Chengyuan, et al. "OpenFWI: Benchmark Seismic Datasets for Machine Learning-Based Full Waveform Inversion." arXiv preprint arXiv:2111.02926 (2021).

[1] Wu, Yue, and Youzuo Lin. "InversionNet: An efficient and accurate data-driven full waveform inversion." IEEE Transactions on Computational Imaging 6 (2019): 419-433.

[1] Zhang, Zhongping, and Youzuo Lin. "Data-driven seismic waveform inversion: A study on the robustness and generalization." IEEE Transactions on Geoscience and Remote sensing 58.10 (2020): 6900-6913.