Callbacks

Callbacks for Anomalib models.

class anomalib.utils.callbacks.CdfNormalizationCallback

Bases: Callback

Callback that standardizes the image-level and pixel-level anomaly scores.

on_predict_batch_end(trainer: Trainer, pl_module: AnomalyModule, outputs: dict, batch: Any, batch_idx: int, dataloader_idx: int) → None

Called when the predict batch ends, normalizes the predicted scores and anomaly maps.

on_test_batch_end(trainer: pl.Trainer, pl_module: AnomalyModule, outputs: STEP_OUTPUT | None, batch: Any, batch_idx: int, dataloader_idx: int) → None

Called when the test batch ends, normalizes the predicted scores and anomaly maps.

on_test_start(trainer: Trainer, pl_module: AnomalyModule) → None

Called when the test begins.

on_validation_batch_end(trainer: pl.Trainer, pl_module: AnomalyModule, outputs: STEP_OUTPUT | None, batch: Any, batch_idx: int, dataloader_idx: int) → None

Called when the validation batch ends, standardizes the predicted scores and anomaly maps.

on_validation_epoch_start(trainer: Trainer, pl_module: AnomalyModule) → None

Called when the validation starts after training.

Use the current model to compute the anomaly score distributions of the normal training data. This is needed after every epoch, because the statistics must be stored in the state dict of the checkpoint file.

setup(trainer: pl.Trainer, pl_module: AnomalyModule, stage: str | None = None) → None

Adds training_distribution metrics to normalization metrics.

class anomalib.utils.callbacks.GraphLogger

Bases: Callback

Log model graph to respective logger.

on_train_end(trainer: Trainer, pl_module: LightningModule) → None

Unwatch model if configured for wandb and log it model graph in Tensorboard if specified.

Parameters:

• trainer – Trainer object which contans reference to loggers.

• pl_module – LightningModule object which is logged.

on_train_start(trainer: Trainer, pl_module: LightningModule) → None

Log model graph to respective logger.

Parameters:

• trainer – Trainer object which contans reference to loggers.

• pl_module – LightningModule object which is logged.

class anomalib.utils.callbacks.ImageVisualizerCallback(task: TaskType, mode: str, image_save_path: str, inputs_are_normalized: bool = True, show_images: bool = False, log_images: bool = True, save_images: bool = True)

Bases: BaseVisualizerCallback

Callback that visualizes the inference results of a model.

The callback generates a figure showing the original image, the ground truth segmentation mask, the predicted error heat map, and the predicted segmentation mask.

To save the images to the filesystem, add the ‘local’ keyword to the project.log_images_to parameter in the config.yaml file.

Visualizer callback.

on_predict_batch_end(trainer: pl.Trainer, pl_module: AnomalyModule, outputs: STEP_OUTPUT | None, batch: Any, batch_idx: int, dataloader_idx: int) → None

Show images at the end of every batch.

Parameters:

• trainer (Trainer) – Pytorch lightning trainer object (unused).

• pl_module (AnomalyModule) – Lightning modules derived from BaseAnomalyLightning object as

• images. (currently only they support logging) –

• outputs (STEP_OUTPUT | None) – Outputs of the current test step.

• batch (Any) – Input batch of the current test step (unused).

• batch_idx (int) – Index of the current test batch (unused).

• dataloader_idx (int) – Index of the dataloader that yielded the current batch (unused).

on_test_batch_end(trainer: pl.Trainer, pl_module: AnomalyModule, outputs: STEP_OUTPUT | None, batch: Any, batch_idx: int, dataloader_idx: int) → None

Log images at the end of every batch.

Parameters:

• trainer (Trainer) – Pytorch lightning trainer object (unused).

• pl_module (AnomalyModule) – Lightning modules derived from BaseAnomalyLightning object as currently only they support logging images.

• outputs (STEP_OUTPUT | None) – Outputs of the current test step.

• batch (Any) – Input batch of the current test step (unused).

• batch_idx (int) – Index of the current test batch (unused).

• dataloader_idx (int) – Index of the dataloader that yielded the current batch (unused).

class anomalib.utils.callbacks.LoadModelCallback(weights_path)

Bases: Callback

Callback that loads the model weights from the state dict.

setup(trainer: Trainer, pl_module: AnomalyModule, stage: str | None = None) → None

Call when inference begins.

Loads the model weights from weights_path into the PyTorch module.

class anomalib.utils.callbacks.MetricVisualizerCallback(task: TaskType, mode: str, image_save_path: str, inputs_are_normalized: bool = True, show_images: bool = False, log_images: bool = True, save_images: bool = True)

Bases: BaseVisualizerCallback

Callback that visualizes the metric results of a model by plotting the corresponding curves.

To save the images to the filesystem, add the ‘local’ keyword to the project.log_images_to parameter in the config.yaml file.

Visualizer callback.

on_test_end(trainer: Trainer, pl_module: AnomalyModule) → None

Log images of the metrics contained in pl_module.

In order to also plot custom metrics, they need to have implemented a generate_figure function that returns tuple[matplotlib.figure.Figure, str].

Parameters:

• trainer (pl.Trainer) – pytorch lightning trainer.

• pl_module (AnomalyModule) – pytorch lightning module.

class anomalib.utils.callbacks.MetricsConfigurationCallback(task: TaskType = TaskType.SEGMENTATION, image_metrics: list[str] | None = None, pixel_metrics: list[str] | None = None)

Bases: Callback

Metrics Configuration Callback.

Create image and pixel-level AnomalibMetricsCollection.

This callback creates AnomalibMetricsCollection based on the

list of strings provided for image and pixel-level metrics.

After these MetricCollections are created, the callback assigns these to the lightning module.

Parameters:

• task (TaskType) – Task type of the current run.

• image_metrics (list[str] | None) – List of image-level metrics.

• pixel_metrics (list[str] | None) – List of pixel-level metrics.

setup(trainer: pl.Trainer, pl_module: AnomalyModule, stage: str | None = None) → None

Setup image and pixel-level AnomalibMetricsCollection within Anomalib Model.

Parameters:

• trainer (pl.Trainer) – PyTorch Lightning Trainer

• pl_module (AnomalyModule) – Anomalib Model that inherits pl LightningModule.

• stage (str | None, optional) – fit, validate, test or predict. Defaults to None.

class anomalib.utils.callbacks.MinMaxNormalizationCallback

Bases: Callback

Callback that normalizes the image-level and pixel-level anomaly scores using min-max normalization.

on_predict_batch_end(trainer: Trainer, pl_module: AnomalyModule, outputs: Any, batch: Any, batch_idx: int, dataloader_idx: int) → None

Called when the predict batch ends, normalizes the predicted scores and anomaly maps.

on_test_batch_end(trainer: pl.Trainer, pl_module: AnomalyModule, outputs: STEP_OUTPUT | None, batch: Any, batch_idx: int, dataloader_idx: int) → None

Called when the test batch ends, normalizes the predicted scores and anomaly maps.

on_test_start(trainer: Trainer, pl_module: AnomalyModule) → None

Called when the test begins.

on_validation_batch_end(trainer: Trainer, pl_module: AnomalyModule, outputs: Union[Tensor, Dict[str, Any]], batch: Any, batch_idx: int, dataloader_idx: int) → None

Called when the validation batch ends, update the min and max observed values.

setup(trainer: pl.Trainer, pl_module: AnomalyModule, stage: str | None = None) → None

Adds min_max metrics to normalization metrics.

class anomalib.utils.callbacks.PostProcessingConfigurationCallback(normalization_method: NormalizationMethod = NormalizationMethod.MIN_MAX, threshold_method: ThresholdMethod = ThresholdMethod.ADAPTIVE, manual_image_threshold: float | None = None, manual_pixel_threshold: float | None = None)

Bases: Callback

Post-Processing Configuration Callback.

Parameters:

• normalization_method (NormalizationMethod) – Normalization method. <none, min_max, cdf>

• threshold_method (ThresholdMethod) – Flag indicating whether threshold should be manual or adaptive.

• manual_image_threshold (float | None) – Default manual image threshold value.

• manual_pixel_threshold (float | None) – Default manual pixel threshold value.

setup(trainer: Trainer, pl_module: LightningModule, stage: str | None = None) → None

Setup post-processing configuration within Anomalib Model.

Parameters:

• trainer (Trainer) – PyTorch Lightning Trainer

• pl_module (LightningModule) – Anomalib Model that inherits pl LightningModule.

• stage (str | None, optional) – fit, validate, test or predict. Defaults to None.

class anomalib.utils.callbacks.TilerConfigurationCallback(enable: bool = False, tile_size: int | Sequence = 256, stride: int | Sequence | None = None, remove_border_count: int = 0, mode: str = 'padding', tile_count: int = 4)

Bases: Callback

Tiler Configuration Callback.

Sets tiling configuration from the command line.

Parameters:

• enable (bool) – Boolean to enable tiling operation. Defaults to False.

• tile_size ([int | Sequence]) – Tile size. Defaults to 256.

• stride ([int | Sequence]) – Stride to move tiles on the image.

• remove_border_count (int, optional) – Number of pixels to remove from the image before tiling. Defaults to 0.

• mode (str, optional) – Up-scaling mode when untiling overlapping tiles. Defaults to “padding”.

• tile_count (SupportsIndex, optional) – Number of random tiles to sample from the image. Defaults to 4.

setup(trainer: pl.Trainer, pl_module: pl.LightningModule, stage: str | None = None) → None

Setup Tiler object within Anomalib Model.

Parameters:

• trainer (pl.Trainer) – PyTorch Lightning Trainer

• pl_module (pl.LightningModule) – Anomalib Model that inherits pl LightningModule.

• stage (str | None, optional) – fit, validate, test or predict. Defaults to None.

Raises:

ValueError – When Anomalib Model doesn’t contain Tiler object, it means the model doesn not support tiling operation.

class anomalib.utils.callbacks.TimerCallback

Bases: Callback

Callback that measures the training and testing time of a PyTorch Lightning module.

on_fit_end(trainer: Trainer, pl_module: LightningModule) → None

Call when fit ends.

Prints the time taken for training.

Parameters:

• trainer (Trainer) – PyTorch Lightning trainer.

• pl_module (LightningModule) – Current training module.

Returns:

None

on_fit_start(trainer: Trainer, pl_module: LightningModule) → None

Call when fit begins.

Sets the start time to the time training started.

Parameters:

• trainer (Trainer) – PyTorch Lightning trainer.

• pl_module (LightningModule) – Current training module.

Returns:

None

on_test_end(trainer: Trainer, pl_module: LightningModule) → None

Call when the test ends.

Prints the time taken for testing and the throughput in frames per second.

Parameters:

• trainer (Trainer) – PyTorch Lightning trainer.

• pl_module (LightningModule) – Current training module.

Returns:

None

on_test_start(trainer: Trainer, pl_module: LightningModule) → None

Call when the test begins.

Sets the start time to the time testing started. Goes over all the test dataloaders and adds the number of images in each.

Parameters:

• trainer (Trainer) – PyTorch Lightning trainer.

• pl_module (LightningModule) – Current training module.

Returns:

None