Component Registry System#

PhenoTypic uses Python’s import system as its component registry. Operations, plotters, and dashboards are discoverable by class name, enabling JSON serialization and dynamic pipeline construction.

How It Works#

When ImagePipeline.from_json() encounters an operation like "GaussianBlur", it resolves the class by:

  1. Searching known PhenoTypic modules (phenotypic.enhance, phenotypic.detect, phenotypic.refine, etc.)

  2. Importing the class by name

  3. Instantiating it with the saved parameters

Registering Custom Operations#

Custom operations are automatically discoverable when:

  1. The class is importable from the current Python environment

  2. The class name matches the name stored in the JSON

For operations defined in your own package:

# my_package/my_enhancer.py
from phenotypic.abc_ import ImageEnhancer

class MyCustomEnhancer(ImageEnhancer):
    strength: float = 1.0

    def _operate(self, image):
        ...
        return image

When loading a pipeline that contains MyCustomEnhancer, ensure my_package is installed and importable.

Plot Registry#

Plotters follow a similar pattern — they register through Python’s class system and are accessible via the image.plot accessor. Static figures come from image.plot.<name>(); interactive views are served through the image.plot.dash.<name>() sub-namespace.

Naming Conventions#

  • Operation class names should be descriptive and end with their type: GaussianBlur (enhancer), OtsuDetector (detector), SmallObjectRemover (refiner)

  • Avoid generic names like MyOperation — the class name appears in pipeline JSON files and should be self-documenting