Annotation Toolbox

napari-nD-annotator

A toolbox for annotating objects one by one in nD

License BSD-3 PyPI Python Version tests codecov napari hub

A toolbox for annotating objects one by one in nD.

This plugin contains some tools to make 2D/3D (and technically any dimensional) annotation easier. Main features:

  • auto-filling labels
  • label slice interpolation (geometric mean, RPSV representation)
  • minimal contour segmentation

If the napari-bbox plugin is also installed (see Installation), you can also

  • list objects annotated with bounding boxes
  • visualize selected objects from different projections

This napari plugin was generated with Cookiecutter using @napari's cookiecutter-napari-plugin template.

Installation

You can install napari-nD-annotator via pip:

pip install napari-nD-annotator

The plugin is also available in napari-hub, to install it directly from napari, please refer to plugin installation instructions at the official napari website.

Note: The bounding box and object list functionality requires the napari-bbox Python package. If you want to use bounding boxes, install napari-bbox separately either using pip or directly from napari. You can also install it together with this plugin:

pip install napari-nD-annotator[all]

If any problems occur during installation or while using the plugin, please file an issue.

Usage

You can start napari with the plugin's widgets already opened as:

napari -w napari-nD-annotator "Object List" "Annotation Toolbox"

Bounding boxes

The main idea is to create bounding boxes around objects we want to annotate, crop them, and annotate them one by one. This has mainly two advantages when visualizing in 3D:

  1. We don't have to load the whole data into memory
  2. The surrounding objects won't occlude the annotated ones, making it easier to check the annotation.

Bounding boxes can be created from the Object list widget. The dimensionality of the bounding box layer will be determined from the image layer. As bounding boxes are created, a small thumbnail will be displayed.

The proposed pipeline goes as follows:

  1. Create a bounding box layer
  2. Select data parts using the bounding boxes
  3. Select an object from the object list
  4. Annotate the object
  5. Repeat from 3.

Slice interpolation

The Interpolation tab contains tools for estimating missing annotation slices from existing ones. There are multiple options:

  • Geometric: the interpolation will be determined by calculating the average of the corresponding contour points.
  • RPSV: A more sophisticated average contour calculation, see the preprint here.
  • Distance-based: a signed distance transform is applied to the annotations. The missing slices will be filled in using their weighted sum.

Note: Geometric and RPSV interpolation works only when there's a single connected mask on each slice. If you want to interpolate disjoint objects (e.g. dividing cells), use distance based interpolation instead.

Note: Distance-based interpolation might give bad results if some masks are too far away from each other on the same slice and there's a big offset compared to the other slice used in the interpolation. If you get unsatisfactory results, try annotating more slices (skip less frames).

Minimal contour

This plugin can estimate a minimal contour, which is calculated from a point set on the edges of the object, which are provided by the user. This contour will follow some kind of image feature (pixels with high gradient or high/low intensity). Features:

  • With a single click a new point can be added to the set. This will also extend the contour with the curve shown in red
  • A double click will close the curve by adding both the red and gray curves to the minimal contour
  • When holding Shift, the gray and red highlight will be swapped, so the other curve can be added to the contour
  • With the Ctrl button down a straight line can be added instead of the minimal path
  • If the anchor points were misplaced, the last point can be removed by right-clicking, or the whole point set can be cleared by pressing Esc
  • The Param value at the widget will decide, how strongly should the contour follow edges on the image. Higher value means higher sensitivity to image data, while a lower value will be closer to straight lines.
  • Different features can be used, like image gradient or pixel intensities, and also user-defined features (using Python)
    • the image is accessed as the image variable, and the features should be stored in the features variable in the small code editor widget

This functionality can be used by selecting the Minimal Contour tab in the Annotation Toolbox widget, which will create a new layer called Anchors.

Warning: Do not remove the Anchors layer!

Warning: Some utility layers appear in the layer list when using the plugin. These are marked with a lock (:lock:) symbol. Do not remove them or modify their data, as this will most probably break the plugin! However, you can change their appearance, e.g. their color settings.

Intensity-based:

Gradient-based:

Custom feature:

Shortcuts

ActionMouseKeyboard
Increment selected labelShift + Wheel ⬆️E
Decrement selected labelShift + Wheel ⬇️Q
Previous sliceCtrl + Wheel ⬆️*A
Next sliceCtrl + Wheel ⬇️*D
Increase paint brush size of labels layerAlt + Wheel ⬆️W
Decrease paint brush size of labels layerAlt + Wheel ⬇️S
Interpolate-Ctrl+I
Change between 'Anchors' and the labels layer-Ctrl+Tab
Jump to layer #i-Ctrl+'i'**

*Built-in functionality of napari

**i: 0-9

Note: you can check the list of available shortcuts by clicking the ? button in the bottom right corner of the main widget.

License

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

Issues

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

Version:

  • 0.1.0

Last updated:

  • 09 May 2023

First released:

  • 01 June 2022

License:

Supported data:

  • Information not submitted

Plugin type:

GitHub activity:

  • Stars: 22
  • Forks: 1
  • Issues + PRs: 12

Python versions supported:

Operating system:

Requirements:

  • numpy
  • magicgui
  • qtpy
  • opencv-python
  • matplotlib
  • napari (>=0.4.11)
  • scikit-image (>=0.19)

Sign up to receive updates