Segment Anything for Microscopy

Segment Anything for Microscopy adapts Meta AI's Segment Anything to biomedical images, with models fine-tuned on microscopy data. It's for researchers who need to label cells or structures such as mitochondria in electron microscopy volumes. The software runs locally on Windows, macOS and Linux and is open source under the MIT license.

Its napari plugin combines automatic segmentation with interactive annotation. Point and box prompts let you identify objects and refine their boundaries, while automatic segmentation can label objects across an image or a full 3D volume. An image series tool handles collections of 2D images or volumes without reopening the annotator for each file.

Tracking covers 2D time series, including cell division. You can follow objects interactively or use automatic tracking through Trackastra. It doesn't support tracking volumetric data over time.

For labs that need models suited to their own images, it includes fine-tuning through a graphical interface and a Python library. The library also makes segmentation available to analysis scripts. Its microscopy models are available through BioImage.IO and work in other tools that support Segment Anything.

It can use a CPU or an NVIDIA GPU. Volumetric segmentation and automatic tracking can take a long time without GPU acceleration.

Install the Python package with pip or conda in a napari environment, then download the required pretrained models. Model terms are separate from the MIT code licence. Older standalone CPU installers are deprecated.