Promptable segmentation models can turn a single 3D bounding box into an instance mask, but placing those boxes quickly becomes a bottleneck when microscopy volumes contain thousands of objects. The new napari plugin napari-TurboBox addresses this by keeping one shared copy of all boxes across every synchronized viewer, so edits propagate without per-viewer state duplication.

In benchmarks with a four-viewer layout, napari-TurboBox updates the three slice viewers in a few milliseconds per mouse move even at 10,000 boxes. The existing napari-bbox plugin, by contrast, exceeds a 33 ms interaction budget beyond roughly 1,000 boxes and takes 159 ms at 10,000 boxes.

The plugin keeps boxes axis-aligned and exports them as per-slice 2D boxes in COCO and YOLO formats. It is implemented as a pure-Python plugin for napari 0.6.6 to 0.9, released under the Apache 2.0 licence, with version 0.1.0 archived on Zenodo. The paper is a preprint on arXiv and has not yet undergone peer review.