Inversion with VAE and a new stochastic gradient descent.

This is the code associated to the manuscript Deep generative models in inversion: a review and development of a new approach based on a variational autoencoder which may be found here. An environment.yml file is provided and may be used as:

conda env create -f environment.yml

to install the needed dependencies. Note that if you use pytorch with a GPU it sometimes works better to use the -c pytorch channel.

Once the dependencies are installed, you may install the code and run the synthetic tests in the manuscript available in each of the jupyter notebooks (.ipynb). The notebooks DGM_inv_linear_comparison.ipynb and DGM_inv_nonlinear_comparison.ipynb run the inversions once the DGMs are trained (they read the trained VAE from a parameter file with extension '.pth').

A brief explanation of contents:

• SGD_DGM.py : main module to run inversion with DGMs. This module defines and then takes inputs from a 'SGDsetup' object.
• test_models : models files for testing (as .npy).
• toy_problems : jupyter notebooks for illustrative 'toy' problems. The notebook toy_SGD_deform.ipynb contains the example for testing the performance of SGD with 'ring' regularization considering a misfit function with three local minima. The notebook toy_VAE_eight.ipynb contains the example of an 'eight-shaped' manifold being approximated with a VAE using different values of alpha and beta.
• SGAN : files needed to run comparison with SGAN and taken from here. In this way, the SGAN is already trained and its parameters come from the '.pth' file.
• VAE : files for training and generation of VAE in proposed approach. The jupyter notebook training_VAE.ipynb is used to train the VAE. A GPU is best for lower computational time in this part. The saved parameters for the VAE trained with alpha=0.1 and beta=1000 are provided for testing inversion directly.