Deepinv

DeepInverse, distributed as the deepinv package, is an open source Python library built on PyTorch for solving imaging inverse problems with deep learning. Inverse problems cover tasks such as denoising, deblurring, inpainting and reconstruction, where a clean image has to be recovered from degraded or indirect measurements. The library is part of the official PyTorch Ecosystem and aims to make research across imaging domains faster and more reproducible through a common modular framework of problems and algorithms. It is aimed at imaging researchers, computational scientists and machine learning engineers. The feature set includes a large collection of predefined imaging operators, deep neural networks with pretrained reconstruction models and denoisers ready to use, and frameworks for plug and play restoration, optimisation and unfolded architectures. It also ships training losses designed for inverse problems, sampling algorithms and diffusion models for uncertainty quantification, and a framework for building datasets. Installation works through pip, conda, uv or pixi on Python 3.10 or higher, with optional extras for datasets, denoisers and physics models, and a nightly build is available from GitHub for anyone who wants the latest changes. DeepInverse is free, open source software. There is no paid tier, hosted service or published pricing; the project offers documentation, a five minute quickstart, worked examples, a user guide and a mailing list for release announcements. The documentation describes the library as under active development, so interfaces can change between releases. Teams building on it should pin a version rather than track the nightly build.