napari-mask-curator 0.1.1

Manual curation in napari: quickly pick the best cell/nuclei masks from segmentation outputs (Cellpose/Baysor/parameter sweeps), with new IDs, undo, overlap pruning, compact save, and fast logging.

A napari dock widget for manual curation of segmentation masks. It is designed for workflows where you generate multiple candidate masks (e.g., Cellpose/Baysor/parameter sweeps), and then quickly pick the best instance per cell/nucleus inside napari with full traceability.

Key features

• Follow → Work: create a working copy *.mask__work from your selected mask layer, so the original stays untouched.
• Collect (Always NEW ID): click an object in __work to copy it into manual_best with a new unique label ID.
• Undo (U / Alt-U): revert the last collect and restore the exact pre-click state.
• Overlap prune (P / Alt-P): remove candidate objects in __work that overlap too much with manual_best (avoid duplicates).
• Compact save (S / Alt-S): before saving, relabel manual_best to consecutive IDs (1..N).
• Fast logging: JSONL + TSV logs of every action (for methods papers / traceability).
• Blink: toggle selected layer visibility at 0.5s to compare masks quickly.

Compatibility

• Python: 3.10–3.13 (tested on 3.11)
• napari: 0.6.x (tested on 0.6.6)

What input masks should look like

• Input mask should be a Labels layer.
• The layer name is expected to contain .mask (example: d290_cp0_fl0.2.mask).
  • This is how the plugin identifies which layers are “candidate masks”.

Tip: if your mask is a LZW-compressed TIFF and napari fails to open it, install imagecodecs:

conda install -c conda-forge imagecodecs

Typical workflow

1. Open napari
2. Load:
   • A background image (optional), e.g. DAPI
   • A candidate mask Labels layer, e.g. d290_cp0_fl0.2.mask
3. Open the panel: Plugins → Mask Curator → Mask Curation
4. Select your *.mask layer and click Follow → Work + Prune
   • Creates *.mask__work
   • Creates/uses manual_best
5. Click Collect: OFF (A) to switch it to Collect: ON
6. Left-click objects in __work to collect into manual_best
7. If you mis-click, press Undo (or U / Alt-U)
8. Press Prune Now (or P / Alt-P) to prune overlaps
9. Press Save manual_best (or S / Alt-S) to export final curated masks with compact IDs.

Hotkeys

• Alt-A: Toggle Collect ON/OFF
• Alt-U: Undo last collect
• Alt-P: Prune Now
• Alt-R: Reset Work
• Alt-S: Save manual_best

(Buttons are also provided for all actions.)

Parameters (panel)

• Overlap prune thr: candidates are removed if overlap ratio ≥ thr
• ROI pad(px): padding around object bbox used for fast local updates

Outputs

When saving, the plugin writes:

• manual_best_ID_uint32.tif — curated Labels with compact consecutive IDs (1..N)
• manual_best_black_uint8.tif — binary mask (0/255)
• manual_best_id_map_old_to_new.tsv — mapping from pre-save IDs to compact IDs

Logs (traceability):

• curation_logs/curation_<timestamp>.jsonl
• curation_logs/curation_<timestamp>.tsv

Installation

From PyPI

pip install napari-mask-curator