ConvDO

Convolutional Differential Operators for Physics-based Deep Learning Study

Calculate the spatial derivative differentiablly!

[📖 Documentation & Examples]

Installation

• Install through pip: pip install git+https://github.com/qiauil/ConvDO
• Install locally: Download the repository and run ./install.sh or pip install .

Feature

Positive😀 and negative🙃 things are all features...

• PyTorch-based and only supports 2D fields at the moment.
• Powered by convolutional neural network.
• Differentiable and GPU supported (why not? It's PyTorch based!).
• Second order for Dirichlet and Neumann boundary condition.
• Up to 8th order for periodic boundary condition.
• Obstacles inside of the domain is supported.

Documentations

Check 👉 here

Further Reading

Projects using ConvDO:

• Diffusion-based-Flow-Prediction: Diffusion-based flow prediction (DBFP) with uncertainty for airfoils.
• To be updated...

If you need to solve more complex PDEs using differentiable functions, please have a check on

• PhiFlow: A differentiable PDE solving framework for machine learning
• Exponax: Efficient Differentiable n-d PDE solvers in JAX.

For more research on physics based deep learning research, please visit the website of our research group at TUM.