frameworks.ultralytics.services.model.callbacks

callbacks

Classes:

Name	Description
UltralyticsCallbacks	Provides callback hooks for training and evaluation lifecycle events in Ultralytics YOLO models.

Functions:

Name	Description
extract_column_names	Parses a YOLO-style descriptor string to extract column names, preserving prefixes like 'Box' and 'Mask'.
log_confusion_matrix

UltralyticsCallbacks(experiment, logger, metric_mapping, model, save_period)

Provides callback hooks for training and evaluation lifecycle events in Ultralytics YOLO models.

This class integrates with Picsellia to log metrics, images, and artifacts during the training and validation process.

Parameters:

Name	Type	Description	Default
experiment	Experiment	The experiment object used for logging.	required
logger	type[TBaseLogger]	The logger class used for metric/image logging.	required
metric_mapping	TMetricMapping	Maps framework-specific metric names to standard names.	required
model	UltralyticsModel	The model wrapper to manage artifacts and paths.	required
save_period	int	Frequency (in epochs) to save model checkpoints.	required

Methods:

Name	Description
on_train_epoch_end	Callback called at the end of each training epoch.
on_fit_epoch_end	Callback called at the end of each epoch (after validation).
on_val_end	Callback called after validation.
on_train_end	Callback called at the end of training.
get_callbacks	Returns the dictionary of callback methods for integration into the Ultralytics engine.

Attributes:

Name	Type	Description
experiment
model
save_period

experiment = experiment instance-attribute

model = model instance-attribute

save_period = save_period instance-attribute

on_train_epoch_end(trainer)

Callback called at the end of each training epoch. Logs training and validation losses, learning rates, and periodically saves the model.

Parameters:

Name	Type	Description	Default
trainer	TBaseTrainer	The trainer object with training state and losses.	required

on_fit_epoch_end(trainer)

Callback called at the end of each epoch (after validation). Logs time, metrics, and training images.

Parameters:

Name	Type	Description	Default
trainer	TBaseTrainer	The trainer object with metrics and epoch info.	required

on_val_end(validator)

Callback called after validation. Logs validation result images and performance metrics.

Parameters:

Name	Type	Description	Default
validator	TBaseValidator	The validator object with validation metrics.	required

on_train_end(trainer)

Callback called at the end of training. Currently logs final metrics (disabled by default).

Parameters:

Name	Type	Description	Default
trainer	TBaseTrainer	The trainer object.	required

get_callbacks()

Returns the dictionary of callback methods for integration into the Ultralytics engine.

Returns:

Name	Type	Description
dict	dict[str, Callable]	Callback function names mapped to their handlers.

extract_column_names(desc)

Parses a YOLO-style descriptor string to extract column names, preserving prefixes like 'Box' and 'Mask'.

Parameters:

Name	Type	Description	Default
desc	str	A description string (e.g., "Class Images Instances Box(P R mAP50)").	required

Returns:

Type	Description
list[str]	list[str]: Parsed column names (e.g., ["Class", "Images", "Instances", "Box(P)", ...]).

log_confusion_matrix(validator, logger)