napari-bootstrapper 0.3.0

A plugin to quickly generate dense ground truth with sparse labels

napari-bootstrapper

A napari plugin to quickly generate dense 3D segmentations from sparse 2D labels.

Sparse 2D annotations made in ~10 minutes on a single section can produce dense 3D segmentations that are good starting points for refinement. Based on the 2D→3D method described in Sparse Annotation is Sufficient for Bootstrapping Dense Segmentation.

For larger volumes, dedicated 3D models, and block-wise processing, see the Bootstrapper CLI.

Dataset	Data Type	Video
CREMI C	3D volumetric stack
Fluo-C2DL-Huh7	2D + time series

---

Getting Started

Install

conda create -n napari-bootstrapper -c conda-forge python==3.11 napari pyqt
conda activate napari-bootstrapper
pip install napari-bootstrapper

Or install the latest development version:

pip install git+https://github.com/ucsdmanorlab/napari-bootstrapper.git

Launch

conda activate napari-bootstrapper
napari

Open the Bootstrapper widget from Plugins → napari-bootstrapper.

---

How It Works

The plugin has four sections that follow a simple workflow:

1. Data

Load a 3D image (or 4D with a channels dimension) and create sparse 2D labels on one or a few slices. Click "Make mask" to generate a binary training mask.

We recommend using a foundation model to make sparse 2D labels, like micro-sam.

2. Train a 2D Model

Train a 2D model on your sparse labels. Three task types are available:

• 2d_affs — affinities
• 2d_lsd — local shape descriptors
• 2d_mtlsd — multi-task (both)

Or load a pretrained checkpoint.

3. Load a 3D Model

The 3D model lifts 2D predictions into 3D affinities. Pretrained weights are recommended — just click "Download".

4. Segment

Click "Start" to run the full 2D→3D inference pipeline. The output is an instance segmentation produced via mutex watershed or connected components.

Proofreading

We provide a separate widget for refining segmentations quickly. Select labels by placing points on them or entering label IDs manually. Operations can be applied per-slice (2D) or on the full volume (3D).

• Morphology — Dilate, erode, open, close, fill holes. Stenciled (3×3×3) or spherical (variable radius). Uses fastmorph.
• Merge / Split — Merge labels, split with watershed markers, or delete. Uses fastremap.
• Filter — Remove by size (min/max voxels), keep K largest, remove outliers by sigma, filter by Z-slice count, relabel connected components. Uses cc3d.

---

Citation

If you find this useful, please cite our preprint:

@article {Thiyagarajan2024.06.14.599135,
	author = {Thiyagarajan, Vijay Venu and Sheridan, Arlo and Harris, Kristen M. and Manor, Uri},
	title = {Sparse Annotation is Sufficient for Bootstrapping Dense Segmentation},
	year = {2024},
	doi = {10.1101/2024.06.14.599135},
	URL = {https://www.biorxiv.org/content/10.1101/2024.06.14.599135v2},
}

Issues

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

Acknowledgements

• micro-sam — Making foundation models like SAM accessible to the community.
• empanada-napari — Proofreading widget inspiration.
• napari-cellulus — Development scaffolding.

Funding

Chan-Zuckerberg Imaging Scientist Award DOI https://doi.org/10.37921/694870itnyzk from the Chan Zuckerberg Initiative DAF, an advised fund of Silicon Valley Community Foundation (funder DOI 10.13039/100014989).

NSF NeuroNex Technology Hub Award (1707356), NSF NeuroNex2 Award (2014862)