RAPIDS cuCIM: porting scikit-image code to the GPU



cuCIM is a new RAPIDS library for accelerated n-dimensional image processing and image I/O. The project is now publicly available under a permissive license (Apache 2.0) and welcomes community contributions. This is the second part of a joint blog post with NVIDIA. Both posts feature a common motivation section, but the NVIDIA post focuses on cuCIM software architecture, image I/O functionality, and benchmark results. In this post, we expand on the CuPy-based cucim.skimage package, which provides a CUDA-based implementation of the scikit-image API. We will give an overview of how existing CPU-based scikit-image code can be ported to the GPU. We will also provide guidance on how to get started using and contributing to cuCIM. The initial release of the library was a collaboration between Quansight and NVIDIA's RAPIDS and Clara teams.

As a quick first look, the following code applies a set of filters that highlight vessels in an image of the retina.

Python source code demonstrating image filtering on the GPU