The POPSICLE Bacterial Segmentation Dataset: multi-class compartment segmentations for bacterial cryo-ET data

Deposition Overview

The bacterial segmentation dataset of POPSICLE (Particle/Object Picking & Segmentation In CryoET Learning & Evaluation), a benchmark suite released to support reproducible evaluation of machine-learning models for dense voxel-wise compartment segmentation in bacterial cryo-ET. The deposition provides semantic segmentations for five classes (cytoplasm, bacterial-type flagellum, membrane, dense body, and periplasmic space / intermembrane space) together with matching 20 Å re-binned WBP tomograms for 80 cryo-ET acquisitions of bacteria drawn from 13 datasets spanning multiple species. Initial annotations were produced semi-manually using napari-nnInteractive, then refined through an agent-designed copick curation pipeline (copick-utils + copick-mcp), followed by manual inspection and correction. Only the final curated segmentation is ingested in this deposition. The accompanying 20 Å tomograms are the resampled volumes used for segmentation (WBP, IMOD reconstruction pipeline; alignment via RAPTOR per the original pipeline) and serve as the visualization base for the segmentations at this voxel spacing.