frameworks.ultralytics.services.model.logger.segmentation

segmentation

Classes:

Name	Description
UltralyticsSegmentationMetricMapping	Defines the metric mapping for segmentation tasks using the Ultralytics framework.
UltralyticsSegmentationLogger	Logger for Ultralytics-based segmentation models.

UltralyticsSegmentationMetricMapping()

Bases: UltralyticsObjectDetectionMetricMapping

Defines the metric mapping for segmentation tasks using the Ultralytics framework.

This class extends the object detection mapping and adds segmentation-specific metrics, such as segmentation loss and mask-based precision, recall, and mAP.

Adds metrics for training and validation segmentation loss, as well as mask precision, recall, mAP50, and mAP50-95.

Methods:

Name	Description
add_metric	Add a metric to the specified phase.
get_mapping	Get mapping of framework names to standard names for a given phase.

Attributes:

Name	Type	Description
mappings	dict[str, list[Metric]]

mappings = {'train': [], 'val': [], 'test': []} instance-attribute

add_metric(phase, metric)

Add a metric to the specified phase.

Parameters:

Name	Type	Description	Default
phase	str	One of 'train', 'val', or 'test'.	required
metric	Metric	The metric to register.	required

get_mapping(phase=None)

Get mapping of framework names to standard names for a given phase.

Parameters:

Name	Type	Description	Default
phase	str | None	One of 'train', 'val', 'test'.	None

Returns:

Name	Type	Description
dict	dict[str, str]	Mapping of metric names.

UltralyticsSegmentationLogger(experiment, metric_mapping)

Bases: BaseLogger

Logger for Ultralytics-based segmentation models.

This logger uses an UltralyticsSegmentationMetricMapping to log metrics during training and validation phases to a Picsellia experiment.

Parameters:

Name	Type	Description	Default
experiment	Experiment	The experiment used for logging.	required
metric_mapping	UltralyticsSegmentationMetricMapping	The segmentation-specific metric mapping.	required

Methods:

Name	Description
log_metric	Log a metric value (e.g. for line plot).
log_value	Log a scalar value (e.g., accuracy score).
log_image	Log an image file.
log_confusion_matrix	Log a confusion matrix as a heatmap.
log_table	Log a table (either a key-value dict or 2D matrix).
get_log_name	Construct log name with optional phase and mapped name.

Attributes:

Name	Type	Description
experiment
metric_mapping

experiment = experiment instance-attribute

metric_mapping = metric_mapping instance-attribute

log_metric(name, value, log_type=LogType.LINE, phase=None)

Log a metric value (e.g. for line plot).

Parameters:

Name	Type	Description	Default
name	str	Metric name.	required
value	float	Metric value.	required
log_type	LogType	Logging type (default LINE).	LINE
phase	str | None	Phase name.	None

log_value(name, value, phase=None, precision=4)

Log a scalar value (e.g., accuracy score).

Parameters:

Name	Type	Description	Default
name	str	Metric name.	required
value	float	Value to log.	required
phase	str | None	Phase name.	None
precision	int	Decimal precision to round.	4

log_image(name, image_path, phase=None)

Log an image file.

Parameters:

Name	Type	Description	Default
name	str	Log name.	required
image_path	str	Path to image.	required
phase	str | None	Phase name.	None

log_confusion_matrix(name, labelmap, matrix, phase=None)

Log a confusion matrix as a heatmap.

Parameters:

Name	Type	Description	Default
name	str	Log name.	required
labelmap	dict	Mapping of label indices to names.	required
matrix	ndarray	Confusion matrix.	required
phase	str | None	Phase name.	None

log_table(name, data, phase=None)

Log a table (either a key-value dict or 2D matrix).

Parameters:

Name	Type	Description	Default
name	str	Log name.	required
data	dict	Data to log.	required
phase	str | None	Phase name.	None

get_log_name(metric_name, phase=None)

Construct log name with optional phase and mapped name.

Parameters:

Name	Type	Description	Default
metric_name	str	Base metric name.	required
phase	str | None	Optional phase.	None

Returns:

Name	Type	Description
str	str	Full log name.