PhyCV is an open-source computer vision library from UCLA’s Jalali Lab, hosted on GitHub with about 530 stars and updates in 2024. It describes itself as the first physics-inspired computer vision library: instead of hand-crafted empirical rules, its algorithms emulate how light propagates through a diffractive medium followed by coherent detection.
The library implements a family of algorithms derived from the Phase-Stretch Transform. The Phase-Stretch Transform (PST) performs edge detection by emulating light propagation through a medium with engineered diffractive properties. PAGE (Phase-Stretch Adaptive Gradient-field Extractor) extends the idea to adaptive edge extraction, and VEViD tackles low-light image and video enhancement using virtual diffraction and coherent detection. PhyCV is distributed on PyPI, comes with Colab notebooks and sample results, and has a companion CUDA-accelerated C++ implementation in a separate repository. The underlying research is described in an arXiv paper from the lab.
The physics angle is more than a metaphor. Where conventional algorithms chain together hand-crafted empirical rules, PhyCV’s transforms are derived from how a physical optical system actually processes an image, which makes their behavior principled and their parameters interpretable. The team presents sample results across edge detection and enhancement tasks, and the library’s algorithms are aimed at fast, high-quality processing that can also run at the edge.
PhyCV is free and open source, distributed through GitHub and PyPI at no cost. There are no paid tiers or hosted services; you run the algorithms in your own environment.
It suits computer vision researchers, imaging scientists, and engineers who want fast, interpretable edge detection and low-light enhancement, and anyone interested in physics-based alternatives to conventional image processing pipelines.