Pyxu (pronounced [piksu], formerly known as Pycsou) is an open-source Python framework allowing scientists at any level to quickly prototype/deploy hardware accelerated and out-of-core computational imaging pipelines at scale. Thanks to its microservice architecture and tight integration with the PyData ecosystem, Pyxu supports a wide range of imaging applications, scales, and computation architectures.
- Universal & Modular 🌐: Unlike other frameworks which are specialized for particular imaging types, Pyxu is a general-purpose computational imaging tool. No more wrestling with one-size-fits-all solutions that don't quite fit!
- Plug-and-Play Functionality 🎮: Craft imaging pipelines effortlessly with advanced operator algebra logic. Pyxu automates the tedious bits, like computing gradients, proximal operators, and computing hyperparameters.
- High-Performance Computing 🚀: Whether you're using CPUs or GPUs, Pyxu works with both. It employs Duck arrays, just-in-time compilation via Numba, and relies on CuPy and Dask for GPU/distributed computing needs.
- Flexible & Adaptable 🛠️: Combat the common woes of software rigidity with Pyxu's ultra-flexible framework. Bayesian techniques requiring extensive software flexibility are a breeze here.
- Hardware Acceleration 🖥️: Leverage built-in support for hardware acceleration to ramp up your computational speed, all thanks to our module-agnostic codebase.
- Distributed Computing 🔗: Got a lot of data? No worries! Pyxu works at scale and is easily deployable on institutional clusters using industry-standard technologies like Kubernetes and Docker.
- Deep Learning Interoperability 🤖: Integrate with major deep learning frameworks like PyTorch and JAX for state-of-the-art computational imaging techniques.
In the realm of computer vision 📷, digital image restoration and enhancement techniques have established themselves as indispensable pillars. These techniques, aiming to restore and elevate the quality of degraded or partially observed images, have witnessed unprecedented progress 📈 in recent times. Thanks to potent image priors, we've now reached an era where image restoration and enhancement methods are incredibly advanced ✨ —so much so that we might be approaching a pinnacle in terms of performance and accuracy.
However, it's not all roses 🌹.
Despite their leaps in progress, advanced image reconstruction methods often find themselves trapped in a vicious cycle of limited adaptability, usability, and reproducibility. Many advanced computational imaging solutions, while effective, are tailored for specific use-cases and seldom venture beyond the confines of a proof-of-concept 🚧. These niche solutions demand deep expertise to customize and deploy, making their adoption in production pipelines challenging.
In essence, the imaging domain is desperately seeking what the deep learning community found in frameworks like PyTorch, TensorFlow, or Keras —a flexible, modular, and powerful environment that accelerates the adoption of cutting-edge methods in real-world settings. Pyxu stands as an answer to this call: a groundbreaking, open-source computational imaging software framework tailored for Python enthusiasts 🐍.
The core of Pyxu is lightweight and straightforward to install. You'll need Python (>= 3.10, < 3.13) and a few
mandatory dependencies. While these dependencies will be automatically installed via pip, we highly recommend
installing NumPy and SciPy via conda to benefit from optimized math libraries.
First, to install NumPy and SciPy from conda-forge:
conda install -c conda-forge numpy scipyAnd then install Pyxu:
pip install pyxuThat's it for the basic installation; you're ready to go! Check out the install guide for instructions on how to build from source, or for more advanced options.
Pyxu offers a comprehensive suite of algorithms, including the latest primal-dual splitting methods for nonsmooth optimization. The feature set is robust and mature, positioning it as a leader in the computational imaging arena.
| Package Name 📦 | Operator Types 🛠️ | Operator Algebra 🎯 | Algorithmic Suite 📚 | Application Focus 🎯 | Remarks 💬 |
|---|---|---|---|---|---|
| PyLops | 🔴 Linear oeprators | 🟡 Partial | 🔴 Least-squares & sparse reconstructions | 🟡 Wave-processing, geophysics | 🔴 Linear operators based on NumPy's old matrix interface |
| PyProximal | 🔴 Proximable functionals | 🔴 None | 🔴 Non-smooth convex optimization | 🟢 None | 🔴 Under early development, unstable API |
| Operator Discretization Library (ODL) | 🟢 (Non)linear operators, differentiable/proximable functionals | 🟢 Full | 🟢 Smooth, non-smooth & hybrid (non-)convex optimization | 🟢 None | 🔴 Domain-specific language for mathematicians |
| GlobalBioIm | 🟢 (Non)linear operators, differentiable/proximable functionals | 🟢 Full | 🟢 Smooth, non-smooth & hybrid convex optimization | 🟢 None | 🔴 MATLAB-based, unlike most DL frameworks |
| SigPy | 🟡 Linear operators, proximable functionals | 🟡 Partial | 🟡 Smooth & non-smooth convex optimization | 🔴 MRI | 🔴 Very limited suite of operators, functionals, and algorithms |
| SCICO | 🟢 (Non)linear operators, differentiable/proximable functionals | 🟢 Full | 🟢 Smooth, non-smooth & hybrid (non-)convex optimization | 🟢 None | 🟡 JAX-based (pure functions only, no mutation, etc.) |
| DeepInv | 🟢 (Non)linear operators, differentiable/proximable functionals | 🟡 Partial | 🟢 Smooth, non-smooth & hybrid (non-)convex optimization | 🟡 Deep Learning | 🟡 PyTorch-based (lots of dependencies) |
| Pyxu | 🟢 (Non)linear operators, differentiable/proximable functionals | 🟢 Full | 🟢 Smooth, non-smooth & hybrid (non-)convex optimization | 🟢 None | 🟢 Very rich suite of operators, functionals, algorithms & HPC features |
Pyxu is unique in supporting both out-of-core and distributed computing. Additionally, it offers robust support for JIT compilation and GPU computing via Numba and CuPy respectively. Most contenders either offer partial support or lack these features altogether.
| Package Name 📦 | Auto Diff/Prox ⚙️ | GPU Computing 🖥️ | Out-of-core Computing 🌐 | JIT Compiling ⏱️ |
|---|---|---|---|---|
| PyLops | 🔴 No | 🟢 Yes (CuPy) | 🔴 No | 🟡 Partial (LLVM via Numba) |
| PyProximal | 🔴 No | 🔴 No | 🔴 No | 🔴 No |
| Operator Discretization Library (ODL) | 🟢 Yes | 🟡 Very limited (CUDA) | 🔴 No | 🔴 No |
| GlobalBioIm | 🟢 Yes | 🟢 Yes (MATLAB) | 🔴 No | 🔴 No |
| SigPy | 🔴 No | 🟢 Yes (CuPy) | 🟡 Manual (MPI) | 🔴 No |
| SCICO | 🟢 Yes | 🟢 Yes + TPU (JAX) | 🔴 No | 🟢 Yes (XLA via JAX) |
| DeepInv | 🟢 Autodiff support | 🟢 Yes (PyTorch) | 🔴 No | 🟡 Partial(XLA via torch.compile) |
| Pyxu | 🟢 Yes | 🟢 Yes (CuPy) | 🟢 Yes (Dask) | 🟢 Yes (LLVM and CUDA via Numba) |
Ready to dive in? 🏊♀️ Our tutorial kicks off with an introductory overview of computational imaging and Bayesian reconstruction. Our user guide then provides an in-depth tour of Pyxu's multitude of features through concrete examples.
So, gear up to embark on a transformative journey in computational imaging.
Pyxu is open-source and ever-evolving 🚀. Your contributions, whether big or small, can make a significant impact. So come be a part of the community that's setting the pace for computational imaging 🌱.
Let's accelerate the transition from research prototypes to production-ready solutions. Dive into Pyxu today and make computational imaging more powerful, efficient, and accessible for everyone! 🎉
@software{pyxu-framework,
author = {Matthieu Simeoni and
Sepand Kashani and
Joan Rué-Queralt and
Pyxu Developers},
title = {pyxu-org/pyxu: pyxu},
publisher = {Zenodo},
doi = {10.5281/zenodo.4486431},
url = {https://doi.org/10.5281/zenodo.4486431}
}