Small forward modelling test
Builds a two-material 2D model, creates a Ricker source, records a receiver waveform, and backpropagates a scalar loss to the relative-permittivity model.
Repository notebooks
Examples grounded in the source.
Three current notebooks cover the path from a minimal forward solve to 2D and 3D dual-parameter inversion.
2D full-waveform inversion
Configures a multi-shot acquisition, optimizes relative permittivity and conductivity with PyTorch, and applies total-variation regularization to a 2D subsurface model.
3D full-waveform inversion
Defines a 3D survey with multiple shots and receivers, reconstructs relative permittivity and conductivity, and visualizes central slices of the model.
Repository model arrays
The examples directory also contains 2DFWImodel.npy and OverThrust.npy, which are consumed by the inversion notebooks as model inputs.
Documented outputs
Models and responses included in the repository.
No synthetic website-only results are shown here; each image comes directly from the versioned Fig/ directory.
Forward response
Receiver data and the relative-permittivity gradient from the documented small test.
2D inverted model
Dual-parameter inversion result for relative permittivity and conductivity.
2D true model
Reference relative-permittivity and conductivity models used for the repository comparison.
2D initial model
Initial relative-permittivity and conductivity models before inversion.
3D true model — central slice
Repository-provided reference slice for the documented 3D FWI comparison.
3D initial model — central slice
Repository-provided starting slice before the documented 3D inversion.
3D inverted model — central slice
Repository-provided reconstructed slice from the documented 3D FWI example.
Run locally
Start from the notebooks.
Clone the repository, install the package dependencies, and select a notebook appropriate for the available CPU or CUDA environment.