Prefab Pipelines: Which One for Which Organism#

Prefab pipelines are pre-configured ImagePipeline subclasses that bundle operations, refiners, and measurements into a single ready-to-use workflow. Each is tuned for a specific imaging scenario.

Overview#

Prefab	Organism	Plate type	Key strategy
`HeavyOtsuPipeline`	Yeast, bacteria	96/384-well grid	Multi-stage Otsu with refinement
`HeavyWatershedPipeline`	Yeast, bacteria	Dense grids	Watershed for touching colonies
`RoundPeaksPipeline`	Yeast	Grid plates	Peak detection, lightweight
`HeavyRoundPeaksPipeline`	Yeast	Grid plates	Extended peak detection with refinement
`FilamentousFungiPipeline`	Neurospora, Aspergillus, etc.	Grid plates	BM3D + FilamentousFungiDetector
`GridSectionPipeline`	Any	Pre-tiled sections	Per-section processing

HeavyOtsuPipeline#

Best for: General-purpose colony detection on clean, well-lit plates with uniform backgrounds.

Pipeline steps: GaussianBlur → CLAHE → MedianFilter → SobelFilter → OtsuDetector → MaskOpener → BorderObjectRemover → SmallObjectRemover → MaskFill → GridOversizedObjectRemover → ReduceMultipleGridObjects → GridAligner

Measurements: MeasureShape, MeasureColor, MeasureTexture, MeasureIntensity

When to choose: This is the default starting point. Try it first; switch to a specialized prefab only if results are unsatisfactory.

HeavyWatershedPipeline#

Best for: Plates where colonies grow large enough to touch or overlap. Watershed segmentation separates merged colonies that Otsu would detect as a single object.

When to choose: Colony overlap is visible in the overlay — adjacent colonies sharing a boundary without clear separation.

RoundPeaksPipeline#

Best for: Grid plates with well-separated round colonies. Faster than the Heavy pipelines because it uses fewer refinement steps.

When to choose: Speed matters and colonies are round, well-spaced, and clearly visible.

FilamentousFungiPipeline#

Best for: Filamentous organisms with branching hyphae. Uses BM3D denoising, homomorphic illumination correction, and the specialized FilamentousFungiDetector with Dijkstra reconnection.

When to choose: The organism grows as a mycelial network rather than a compact round colony. Standard detectors miss the outer hyphae.

Customizing Prefabs#

Prefabs expose their key parameters through __init__:

from phenotypic.prefab import HeavyOtsuPipeline

pipeline = HeavyOtsuPipeline(
    gaussian_sigma=3,       # Stronger blur
    small_object_min_size=100,  # Remove larger fragments
    benchmark=True,         # Track execution times
)

For more extensive customization, build a pipeline from scratch using ImagePipeline(ops=[...], meas=[...]).