napari-align-annotator

A napari plugin for building annotated datasets for ALIGN. Annotation work is organised into projects: a directory holding a project.yaml descriptor and a master annotation CSV, so a session can be closed and resumed at any time.

Installation

Install the plugin into an environment that has napari and a Qt binding:

pip install napari-align-annotator

or, to get napari and PyQt together:

pip install "napari-align-annotator[all]"

The plugin can also be installed from napari's plugin manager (Plugins > Install/Uninstall Plugins...).

To install the latest development version:

pip install git+https://github.com/spsalmon/napari-align-annotator.git

Usage

Open napari and start the widget from Plugins > ALIGN Annotator. From there you can either Create Project or Load Project (pick a directory containing a project.yaml).

When creating a project you choose a name, a location, the image type (multichannel, z-stack or time series), the data directories, and optionally whether the data should be copied into the project. Two project types are currently supported:

Classification

Assign one class to each whole image.

• Define the list of classes, and choose whether to display the image, its mask, or both (masks require mask directories).
• Click a class button to label the current image and move to the next one. Ignore removes the current image from the project.
• Labels are written to annotations/annotations.csv, with one row per image (ImagePath, MaskPath, Class).

Panoptic

Assign a class to each segmented instance of an image.

• Requires raw image directories and matching segmentation (label) directories, plus a list of classes.
• Select a class, then place a point on an instance to label it. For z-stacks, instances are linked across planes (stitched by IoU, threshold adjustable with Re-stitch) so a label placed on one plane propagates to the rest of the object.
• Keyboard shortcuts: J next image, H previous image, S save.
• The master CSV lists Reference, Segmentation and Annotation paths; each per-image annotation file has one row per labelled instance (Label, ClassID, Class, plus Z and InstanceID for z-stacks). Label is the value of the instance in the segmentation file.

The Keypoint project type is shown in the creator but is not supported yet.

Contributing

Contributions are very welcome. Tests can be run with tox or directly with pytest after pip install -e ".[testing]"; please ensure the coverage at least stays the same before you submit a pull request.

License

Distributed under the terms of the BSD-3 license, "napari-align-annotator" is free and open source software.

Issues

If you encounter any problems, please file an issue along with a detailed description.

This napari plugin was generated with copier using the napari-plugin-template.