traccuracy: Evaluate Cell Tracking Solutions

traccuracy provides a suite of benchmarking functions that can be used to evaluate cell tracking solutions against ground truth annotations. The goal of this library is to provide a convenient way to run rigorous evaluation and to document and consolidate the wide variety of metrics used in the field.

traccuracy can compute a comprehensive set of metrics for evaluating cell linking and division performance, and can compute biologically meaningful metrics such as the number of correctly reconstructed lineages over N frames and cell cycle length accuracy. As matching ground truth and predicted lineages is a crucial step for performing evaluation, traccuracy includes a number of algorithms for matching ground truth and predicted lineages, both with and without segmentation masks.

Learn more in the documentation or check out the source code.

Installation

pip install traccuracy

How It Works

The traccuracy library has three main components: loaders, matchers, and metrics.

Loaders load tracking graphs from other formats, such as the CTC format, into a TrackingGraph object. A TrackingGraph is a spatiotemporal graph backed by a networkx.DiGraph Nodes represent a single cell in a given time point, and are annotated with a time and a location. Edges point forward in time from a node representing a cell in time point t to the same cell or its daughter in frame t+1 (or beyond, to represent skip edges). Additional terminology is documented in the glossary To load TrackingGraphs from a custom format, you will likely need to implement a loader: see documentation here for more information. Alternatively you can initialize a TrackingGraph with a networkx.DiGraph and ArrayLike objects of segmentation masks if needed.

Matchers take a ground truth and a predicted TrackingGraph with optional segmentation masks and match the nodes and edges to allow evaluation to occur. A list of matchers is available here.

In order to compute metrics, traccuracy begins by annotating the matched graphs with error flags such as False Positive and False Negative. The annotated graph can be exported and used for visualization in other tools. Finally, metrics inspect the error annotations to report both error counts and summary statistics.

The traccuracy library has a flexible Python API, shown in this example notebook. Additionally there is a command line interface for running standard CTC metrics, documented here.

from traccuracy.loaders import load_ctc_data
from traccuracy.matchers import PointMatcher
from traccuracy.metrics import DivisionMetrics, BasicMetrics

# Load data in TrackingGraph objects
gt_data = load_ctc_data(
    "path/to/GT/TRA",
    "path/to/GT/TRA/man_track.txt",
    name="GT"
)
pred_data = load_ctc_data(
    "path/to/prediction",
    "path/to/prediction/track.txt",
    name="prediction"
)

results, matched = run_metrics(
    gt_data=gt_data,
    pred_data=pred_data,
    matcher=PointMatcher(),
    metrics=[DivisionMetrics(), BasicMetrics()]
)

Implemented Metrics

• Cell Tracking Challenge Evaluation Methodology
  • CTC-DET
  • CTC-LNK
  • CTC-TRA
  • Branching Correctness (BC)
• Acyclic Oriented Graph Metric (AOGM) from Matula et al. 2015. A generalized form the CTC metrics where you can supply different weights for each component of the overall metric.
• Division Performance Metrics Optionally allows detection within N frames of ground truth division.
  • Division Precision
  • Division Recall
  • Division F1 score
  • Mitotic Branching Correctness from Ulicna et al. 2021.
• Basic Connectivity Metrics highlighting performance on both nodes and edges.
  • Node/Edge Precision
  • Node/Edge Recall
  • Node/Edge F1 Score
• Track Overlap Metrics defined by Bise et al. 2011, Chen 2021, and Fukai et al. 2022.
  • Track Purity
  • Track Effectiveness
  • Track Fractions
  • Complete Tracks
• CHOTA from Kaiser et al. 2024.

Featured Works

If you use traccuracy in your own work, please let us know so that we can feature it here!

• Archit, A. et al. (2025). Segment anything for microscopy. Nature Methods.
• Evans et al. (2025). Icebergs, jigsaw puzzles and genealogy: Automated multi-generational iceberg tracking and lineage reconstruction. Preprint in EGUsphere.
• Seiffarth, J., & Nöh, K. (2025). PyUAT: Open-source Python framework for efficient and scalable cell tracking. arXiv preprint.
• Bragantini, J. et al. (2024). Ultrack: pushing the limits of cell tracking across biological scales. bioRxiv.
• Gallusser, B., & Weigert, M. (2024). Trackastra: Transformer-based cell tracking for live-cell microscopy. In European Conference on Computer Vision.
• Toma, T. T., Wang, Y., Gahlmann, A., & Acton, S. T. (2024). Deep Temporal Sequence Classification and Mathematical Modeling for Cell Tracking in Dense 3D Microscopy Videos of Bacterial Biofilms. arXiv preprint.
• Lalit, M. & Funke, J. (2025) An Investigation of Unsupervised Cell Tracking and Interactive Fine-Tuning. In International Conference on Computer Vision (ICCV) Workshops