patchworks

Tiled processing of arbitrarily large images — any image, any function.

┌──────┬──────┬──────┐                    ┌──────┬──────┬──────┐
│      │      │      │   fn(tile) → IDs   │  1   │  2   │  3   │
├──────┼──────┼──────┤  ───────────────►  ├──────┼──────┼──────┤
│      │ 2 TB │      │                    │  4   │  5   │  6   │
├──────┼──────┼──────┤                    ├──────┼──────┼──────┤
│      │      │      │                    │  7   │  8   │  9   │
└──────┴──────┴──────┘                    └──────┴──────┴──────┘
         tiles                               globally consistent labels

Segmentation tools assume the image fits in memory; a light-sheet or whole-slide volume does not. patchworks splits it into tiles, runs any segmentation function on each, and stitches the results so that an object crossing a tile boundary keeps one label. The labels are stored with the image in one OME-Zarr, with a table of every object.

Where to start

You want to	Read
Try it on an image from Python	Getting started
Segment a large image on a SLURM cluster, maybe several stains related to each other	Cluster workflow
Run one step from a terminal	Command line
Use your own model or function	Custom segmentation function
Cells from a membrane stain	Membrane cells
Check and correct the results	Reviewing

Building blocks

Each step is a function you can use on its own:

Step	Function
Convert any image to a pyramidal, calibrated OME-Zarr	to_ome_zarr
Segment tile by tile, with a halo for context	tile_process, stage_tile
Stitch tiles into consistent labels	merge_tile_labels
Measure every object; relate two label images	measure_objects, label_relations
View and review in napari	view_in_napari, patchworks review