core.services.model.evaluator.model_evaluator

model_evaluator

Classes:

Name	Description
ModelEvaluator	Evaluates model predictions and logs metrics into a Picsellia experiment.

ModelEvaluator(experiment, inference_type)

Evaluates model predictions and logs metrics into a Picsellia experiment.

Supports classification, detection (rectangle), OCR, and segmentation (polygon) evaluations, including COCO-style and sklearn metrics.

Parameters:

Name	Type	Description	Default
experiment	Experiment	The experiment where results will be logged.	required
inference_type	InferenceType	Type of inference (classification, detection, etc.).	required

Methods:

Name	Description
evaluate	Add and compute evaluation metrics from a list of predictions.
add_evaluation	Add a single prediction to the experiment as evaluation.
compute_coco_metrics	Compute COCO metrics and log them into the experiment.
compute_classification_metrics	Compute sklearn classification metrics (acc, precision, recall, F1).

Attributes:

Name	Type	Description
experiment
inference_type
experiment_logger

experiment = experiment instance-attribute

inference_type = inference_type instance-attribute

experiment_logger = BaseLogger(experiment=experiment, metric_mapping=MetricMapping()) instance-attribute

evaluate(picsellia_predictions)

Add and compute evaluation metrics from a list of predictions.

Parameters:

Name	Type	Description	Default
picsellia_predictions	list	List of PicselliaPrediction objects.	required

add_evaluation(evaluation)

Add a single prediction to the experiment as evaluation.

Parameters:

Name	Type	Description	Default
evaluation	PicselliaClassificationPrediction | PicselliaRectanglePrediction | PicselliaPolygonPrediction | PicselliaOCRPrediction	A prediction (classification, rectangle, OCR, or polygon).	required

compute_coco_metrics(assets, output_dir, training_labelmap)

Compute COCO metrics and log them into the experiment.

Parameters:

Name	Type	Description	Default
assets	list | MultiAsset	Assets to evaluate.	required
output_dir	str	Directory to save metrics.	required
training_labelmap	dict	Label ID-to-name mapping.	required

compute_classification_metrics(assets, output_dir, training_labelmap)

Compute sklearn classification metrics (acc, precision, recall, F1).

Parameters:

Name	Type	Description	Default
assets	list | MultiAsset	Assets to evaluate.	required
output_dir	str	Output directory.	required
training_labelmap	dict	Label ID-to-name mapping.	required