Outputs

baysor run writes to the directory selected by -o (default segmentation). The default output style is legacy; select Parquet output with:

baysor run -m 30 -s 8 --output-style parquet -o out molecules.csv

Files

Contents	Legacy (default)	Parquet
Molecule assignments	segmentation.csv	molecules.parquet
Cell statistics	segmentation_cell_stats.csv	cells.parquet
Joined 2D polygons	segmentation_polygons_2d.json	cell_boundaries.parquet
Per-layer polygons (3D only)	segmentation_polygons_3d.json	cell_boundaries_3d.parquet
Gene × cell count matrix	segmentation_counts.loom or segmentation_counts.tsv	feature_matrix.h5
Config values and invocation	segmentation_params.dump.toml	run_params.toml
Log	segmentation_log.log	run.log

For 3D data, joined polygons pool each cell's molecules across the z-stack. In molecule tables, cell 0 means noise / unassigned; other cells are named cell_1, cell_2, etc. See Output file formats for columns and storage layouts.

Add -p / --plot for two self-contained browser reports in either style:

• diagnostic_report.html — quality checks and run diagnostics
• segmentation_plot.html — molecule and cell visualization

Legacy output

Use legacy output for CSV / GeoJSON / Loom workflows and Xenium Ranger. --count-matrix-format tsv changes the default Loom matrix to TSV. --polygon-format selects FeatureCollection (default), GeometryCollection, GeometryCollectionLegacy or none (no polygon files).

For Xenium Ranger import, keep transcript IDs and choose the polygon format for your Ranger version. Molecule assignments and joined polygons use matching cell IDs; failed boundary estimates get fallback rectangles.

Parquet output

Use Parquet / GeoParquet tables and the 10x-style HDF5 matrix for downstream Python, R or DuckDB analysis. Transcript IDs are not included in this bundle; use legacy output for Xenium Ranger.

--polygon-format and --count-matrix-format are legacy-only and ignored for Parquet output.