SABER — Segment Anything Based Electron Recognition¶
SABER is an open-source platform for autonomous segmentation of organelles in cryo-electron tomography (cryo-ET) and electron microscopy (EM) datasets. It combines state-of-the-art foundational models with expert-driven classification to deliver reliable, scalable segmentations — from a single micrograph to an entire project.
Why SABER?¶
-
Foundational model power
Zero-shot segmentation using SAM2 and SAM3 — no annotations required to get started.
-
Expert-driven accuracy
Train lightweight classifiers on your own annotations to distinguish organelles from contaminants and artifacts.
-
2D and 3D
Segment single micrographs or propagate across full tomographic volumes.
-
Publication-ready output
Export instance segmentations, semantic maps, coordinates, and per-organelle statistics.
Tutorials¶
CLI¶
-
Pre-processing
Prepare your EM/cryo-ET datasets and annotate segmentations using the interactive GUI.
-
Training a Classifier
Train a domain expert classifier on your annotations to filter SAM2 mask proposals.
-
Inference (2D & 3D)
Apply your trained classifier to generate segmentations across entire datasets.
-
Membrane Refinement
Enforce topological consistency across organelle and membrane segmentations.
Python API¶
-
API Overview
Comprehensive introduction to the SABER Python API.
-
2D Quickstart
Segment 2D micrographs programmatically in a few lines of code.
-
3D Quickstart
Segment 3D tomograms and propagate across volumes.
-
Training Guide
Customize the training loop — architectures, loss functions, and data augmentation.
Getting Help¶
Open an issue on our GitHub repository.