POPSICLE: Particle/Object Picking & Segmentation Benchmark for CryoET

POPSICLE is a benchmark suite for cryo-electron tomography covering dense voxel-wise segmentation and sparse localization tasks. The dataset is built directly on the CryoET Data Portal and includes sub-benchmarks for phantom localization, yeast segmentation, bacterial segmentation, and motor protein localization. It was authored by 'biohub' and last updated on 2026-05-06.

Use Cases

• Benchmarking segmentation models based on dense voxel-wise segmentation of cellular structures.
• Evaluating localization algorithms based on sparse localization of macromolecular complexes.
• Training machine learning models for yeast and bacterial segmentation tasks.
• Developing particle picking tools for cryo-electron tomography data analysis.

Strengths

• Covers multiple distinct sub-benchmarks including phantom localization, yeast segmentation, bacterial segmentation, and motor protein localization.
• Built directly on the authoritative CryoET Data Portal, suggesting a foundation in real experimental data.
• Last updated on 2026-05-06, indicating recent maintenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for large-scale training.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

CryoET Data Portal

Collection Method

Likely aggregated and curated from experimental cryo-electron tomography data.

Freshness

Last updated 2026-05-06 21:20:59.