Image Base Datamodule

Base Anomalib data module.

This module provides the base data module class used across Anomalib. It handles dataset splitting, validation set creation, and dataloader configuration.

The module contains:

• AnomalibDataModule: Base class for all Anomalib data modules

Example

Create a datamodule from a config file:

>>> from anomalib.data import AnomalibDataModule
>>> data_config = "examples/configs/data/mvtec.yaml"
>>> datamodule = AnomalibDataModule.from_config(config_path=data_config)

Override config with additional arguments:

>>> override_kwargs = {"data.train_batch_size": 8}
>>> datamodule = AnomalibDataModule.from_config(
...     config_path=data_config,
...     **override_kwargs
... )

class anomalib.data.datamodules.base.image.AnomalibDataModule(train_batch_size, eval_batch_size, num_workers, train_augmentations=None, val_augmentations=None, test_augmentations=None, augmentations=None, val_split_mode=None, val_split_ratio=None, test_split_mode=None, test_split_ratio=None, seed=None)

Bases: LightningDataModule, ABC

Base Anomalib data module.

This class extends PyTorch Lightning’s LightningDataModule to provide common functionality for anomaly detection datasets.

Parameters:

• train_batch_size (int) – Batch size used by the train dataloader.

• eval_batch_size (int) – Batch size used by the val and test dataloaders.

• num_workers (int) – Number of workers used by the train, val and test dataloaders.

• train_augmentations (Transform | None) – Augmentations to apply to the training images Defaults to None.

• val_augmentations (Transform | None) – Augmentations to apply to the validation images. Defaults to None.

• test_augmentations (Transform | None) – Augmentations to apply to the test images. Defaults to None.

• augmentations (Transform | None) – General augmentations to apply if stage-specific augmentations are not provided.

• val_split_mode (ValSplitMode | str | None) –

  Method to obtain validation set. Options:

  • none: No validation set

  • same_as_test: Use test set as validation

  • from_test: Sample from test set

  • synthetic: Generate synthetic anomalies

• val_split_ratio (float | None) – Fraction of data to use for validation

• test_split_mode (TestSplitMode | str | None) –

  Method to obtain test set. Options:

  • none: No test split

  • from_dir: Use separate test directory

  • synthetic: Generate synthetic anomalies

  Defaults to None.

• test_split_ratio (float | None) – Fraction of data to use for testing. Defaults to None.

• seed (int | None) – Random seed for reproducible splitting. Defaults to None.

property category: str

Get dataset category name.

Returns:

Name of the current category

Return type:

str

classmethod from_config(config_path, **kwargs)

Create datamodule instance from config file.

Parameters:

• config_path (str | Path) – Path to config file

• **kwargs – Additional args to override config

Returns:

Instantiated datamodule

Return type:

AnomalibDataModule

Raises:

• FileNotFoundError – If config file not found

• ValueError – If instantiated object is not AnomalibDataModule

Example

Load from config file:

>>> config_path = "examples/configs/data/mvtec.yaml"
>>> datamodule = AnomalibDataModule.from_config(config_path)

Override config values:

>>> datamodule = AnomalibDataModule.from_config(
...     config_path,
...     data_train_batch_size=8
... )

property name: str

Name of the datamodule.

Returns:

Class name of the datamodule

Return type:

str

predict_dataloader()

Get prediction dataloader.

By default uses the test dataloader.

Returns:

Prediction dataloader

Return type:

Any

setup(stage=None)

Set up train, validation and test data.

This method handles the data splitting logic based on the configured modes.

Parameters:

stage (str | None) – Current stage (fit/validate/test/predict). Defaults to None.

Return type:

None

property task: TaskType

Get the task type.

Returns:

Type of anomaly task (classification/segmentation)

Return type:

TaskType

Raises:

AttributeError – If no datasets have been set up yet

test_dataloader()

Get test dataloader.

Returns:

Test dataloader

Return type:

Any

train_dataloader()

Get training dataloader.

Returns:

Training dataloader

Return type:

Any

val_dataloader()

Get validation dataloader.

Returns:

Validation dataloader

Return type:

Any