Didascalie

Didascalie is a desktop application for annotating medical and biomedical images. Drawing, image processing and model training all run on your own machine. Nothing is uploaded.

A lot of medical image data cannot be sent to a cloud service. The tools that work offline usually expect you to maintain a Python environment or to adopt a whole analysis pipeline. Didascalie is an installer and a project file.

The editor: masks and vector shapes on the same image.

Assisted labelling: a rough stroke refined by an operator.

The gallery: filter sequences and track review status. Batch classify sequences

The inspector: play a sequence back with its labels.

Frame registration from keypoint correspondences.

3D volume mode (experimental): a sequence as a volume.

Multiple users can be working on a same project; inspect the intergrader assessment

Status

This is a solo research project under active development. The core annotation workflow is usable day to day. Features marked experimental are less reliable.

Where to start

• Installing

  Installers for Windows, macOS and Linux, plus what you need for GPU training.

• Quickstart

  Create a project, annotate an image, and export the result.

• User guide

  Projects, annotation types, assisted labelling, review workflow.

• Several annotators

  Accounts and roles in one project file, and inter-grader agreement.

• Python library

  Read and write project files from Python, and convert to COCO or YOLO.

What it can annotate

Type	Notes
Segmentation masks	Brush, polygon, line, point, flood fill
Vector shapes	Polygons, lines and keypoints, convertible to and from masks
Classification	Multiclass and multilabel, several tasks per project
Keypoints	Including correspondences between frames for registration
Text notes	Per frame, attached to a configurable text task

Raster and vector annotations are drawn on the same image and share one undo history.

Project files

A project is a single .dida file. It is an ordinary SQLite database that holds the images (embedded or referenced on disk), the annotations of every account, the label definitions, any trained model and the project settings. To back up a project or give it to a colleague, copy that file.

Several annotators

One project file can hold the work of several people. Each has an account, with or without a password, and annotates the same images independently: nobody sees or overwrites anyone else's masks. Accounts are either editors, who annotate, or administrators, who also manage the accounts and the project's labels.

An inter-grader agreement page lets administrators compare the graders pair by pair, with Dice, IoU and kappa scores and pictures of where they differ.

A project used by one person works as it always has, with no account to pick. See Several annotators.