BioImageFlow¶

A Python library for orchestrating bioimage analysis workflows as reproducible DAG pipelines.

BioImageFlow lets you declare image-processing tools, wire them into directed acyclic graphs, and execute them with automatic caching, type checking, and provenance tracking.

Contributor guidance for reusable tool packages, workflow examples, and review checklists lives in BioImageFlow Tool Package Strategy and Agent Tool and Workflow Playbook.

from pathlib import Path
from typing import Annotated

from bioimageflow_core import (
    ProcessingTool, GENERAL_ENV, ImageSpec, Arguments, Template,
)
from bioimageflow import Workflow, configure_wetlands
from bioimageflow_common_tools import Files

class Threshold(ProcessingTool):
    display_name = "Threshold"
    environment = GENERAL_ENV

    class Inputs:
        image: Annotated[Path, ImageSpec()]
        cutoff: float = 128.0

    class Outputs:
        mask: Annotated[Path, ImageSpec(semantics={"binary"})] = Template(
            "{image.stem}_mask.tif"
        )

    def process_row(self, arguments: Arguments) -> "Threshold.Outputs":
        import imageio.v3 as iio
        import numpy as np

        img = iio.imread(arguments.image)
        mask = (img > arguments.cutoff).astype(np.uint8) * 255
        iio.imwrite(arguments.mask, mask)
        return self.Outputs(mask=arguments.mask)

threshold = Threshold()
files = Files()

configure_wetlands(wetlands_instance_path="./wetlands")

with Workflow(storage_path="./bif_data") as wf:
    raw = files(path="/data/images", pattern="*.tif")
    masks = threshold(image=raw["path"], cutoff=100.0, name="threshold")
    result = wf.compute(masks)

Features¶

DAG workflow engine — build pipelines by connecting tools, not writing glue code
Two-package architecture — a minimal worker-safe core, a pandas/pydantic orchestrator for the main process
Typed image I/O — semantic, layout, and dtype constraints checked at graph-construction time
Automatic caching — result-key/current-record caching skips redundant computation
Shared memory — zero-copy array transfer between tools
Merge strategies — inner join, cross join, concat, and collect
Output templating — declarative output path patterns
Environment isolation — each tool declares its own dependencies

GUI / Platform Integrators

GUI / Platform Integrators

Tool Packages

Reference