CITrackastra
OME-Zarr · Bilayers · BIOMERO

From instances
to cell
histories.

Connect segmented cells across time. CITrackastra brings transformer-based tracking to calibrated microscopy images, with traceable trajectories and physical measurements.

Explore actual trajectories ↗Get started locally

Powered by Trackastra · local pretrained weights · research workflow

Fluorescent mouse stem-cell nuclei from the calibrated GOWT1 public dataset
CALIBRATED INPUT. TRACEABLE OUTPUT.PUBLIC MICROSCOPY
3upstream pretrained model variants
2D + timecalibrated HCS plates only
µmcalibrated coordinates and motion
Offlineexplicit local model cache
From labels to lineages

Keep the biology.
Connect the timepoints.

Segment first with CISegmentation or another instance model. Trackastra predicts associations; CITrackastra preserves source pixels and records every link.

01 / INPUT

Bring your instances

OME-Zarr HCS plates with 2D time series and matching labels. Every field is checked independently.

02 / CALIBRATION

Require physical units

Positive XY sizes and matching labels in every field. Volumes and static plates are rejected.

03 / ASSOCIATION

Predict the links

General 2D, CTC or SAM2 features. Greedy linking with or without divisions, or ILP optimization.

04 / OUTPUT

Inspect cell histories

Additional trajectory labels, SQLite and CSV tables, source IDs and model provenance.

Open evidence

See the predictions.
Know their limits.

The gallery uses actual predictions on public, physically calibrated data. Reference markers are inputs for smoke validation; this is not a segmentation benchmark.

01 / Gallery

Play through real trajectories ↗

Move through short and full-length sequences, inspect source images and follow predicted paths.

02 / Usage

One headless workflow ↗

Local CLI, NVIDIA CUDA, cached models and Bilayers/BIOMERO integration.

03 / Evidence

Validation and limitations ↗

Test coverage, actual device runs, calibration sources and deployment status.

Getting started

Begin locally.

Download all three tracking checkpoints and the SAM2 backbone once, then infer from the local cache.

Installation and commands ↗
git clone https://github.com/Cellular-Imaging-Amsterdam-UMC/citrackastra.git
cd citrackastra
create_env.cmd
install_cuda.cmd
.venv\Scripts\python tools\download_models.py

Python 3.11 locally · CPU or NVIDIA CUDA · Docker images and usage.