multimeditron.dataset.loader package

Subpackages

Module contents

class multimeditron.dataset.loader.AutoModalityLoader

Bases: object

Loader registry to automatically manage and instantiate modality loaders.

The AutoModalityLoader acts as a central registry for different modality loader classes, enabling easy registration, instantiation, and retrieval of loaders. Each loader must inherit from BaseModalityLoader to be compatible.

_registry

Internal registry of modality loader classes keyed by name.

Type:

Dict[str, Type[BaseModalityLoader]]

classmethod from_name(name: str, *args, **kwargs) → BaseModalityLoader

Retrieve and instantiate a registered loader class by name.

Parameters:

  • name (str) – The registered name of the loader class.

  • *args – Positional arguments to pass to the loader’s initializer.

  • **kwargs – Keyword arguments to pass to the loader’s initializer.

Returns:

The instantiated loader.

Return type:

BaseModalityLoader

Raises:

ValueError – If the name is not registered.

classmethod register(name: str)

Register a modality loader class under a specific name.

Parameters:

name (str) – The name under which the loader class will be registered.

Returns:

A decorator to register the loader class.

Return type:

Callable

Raises:

ValueError – If the class is not a subclass of BaseModalityLoader or if the name is already registered.

class multimeditron.dataset.loader.BaseModalityLoader

Bases: ABC

Abstract base class for modality loaders.

abstractmethod load(*args, **kwargs) → Any

Abstract method to load modality data. This method must be implemented by subclasses.

Parameters:

  • *args – Variable length argument list.

  • **kwargs – Arbitrary keyword arguments.

Returns:

The loaded modality data.

Return type:

Any

static load_modalities(sample: Dict[str, Any], loaders: Dict[str, BaseModalityLoader])

Load modalities in a sample using the provided modality loaders. This function loads the modality if needed using the loader defined in loaders For instance: If loaders is a dictionary containing an element "image" : FileSystemImageLoader, then this function assumes that every samples store images as path and load the image using the FileSystemImageLoader

Note that each sample should contain a key MODALITY_TYPE_KEY defining what kind of modality this is (for instance “image”). And a key MODALITY_VALUE_KEY defining the value of the modality, this could be a path or the bytes of the images for instance.

Parameters:

  • sample (Dict[str, Any]) – The input sample containing modalities.

  • loaders (Dict[str, BaseModalityLoader]) – A dictionary mapping modality types to their corresponding loaders.

Returns:

The processed sample with merged modalities.

Return type:

Dict[str, Any]

Raises:

ValueError – If a modality type is not found in the loaders.

name: str
class multimeditron.dataset.loader.FileSystemImageLoader(base_path: str | Path)

Bases: BaseModalityLoader

Loader for image files from the filesystem. Expects the sample dictionary to have a “value” key containing the relative path to the image file. The base_path parameter specifies the root directory where images are stored. Example:

loader = FileSystemImageLoader(base_path="/path/to/images")
sample = {"value": "image1.jpg", "type": "image"}
image = loader.load(sample)
# image is a PIL Image object
load(sample: Dict[str, Any]) → Image

Load an image from the filesystem. :param sample: A dictionary containing at least the “value” key with the relative path to the image file. :type sample: Dict[str, Any]

Returns:

The loaded image as a PIL Image object.

Return type:

PIL.Image.Image

name: str = 'fs-image'
class multimeditron.dataset.loader.RawImageLoader

Bases: BaseModalityLoader

Loader for raw image bytes. Expects the sample dictionary to have a “value” key containing a dictionary with a “bytes” key holding the raw image bytes. Example:

loader = RawImageLoader()
sample = {"value": {"bytes": b'...'}, "type": "image"}
image = loader.load(sample)
# image is a PIL Image object
load(sample: Dict[str, Any]) → Image

Load an image from raw bytes.

Parameters:

sample (Dict[str, Any]) – A dictionary containing at least the “value” key with a dictionary that has a “bytes” key holding the raw image bytes.

Returns:

The loaded image as a PIL Image object.

Return type:

PIL.Image.Image

name: str = 'raw-image'