turi::object_detection::average_precision_calculator Class Reference

#include <toolkits/object_detection/od_evaluation.hpp>

Public Member Functions
	average_precision_calculator (flex_list class_labels, std::vector< float > iou_thresholds)
void	add_row (const std::vector< neural_net::image_annotation > &predictions, const std::vector< neural_net::image_annotation > &ground_truth)
variant_map_type	evaluate ()

Detailed Description

Helper class for computing AP and mAP metrics.

Definition at line 34 of file od_evaluation.hpp.

Constructor & Destructor Documentation

average_precision_calculator()

turi::object_detection::average_precision_calculator::average_precision_calculator	(	flex_list	class_labels,
		std::vector< float >	iou_thresholds
	)

Parameters

class_labels	Each prediction and ground truth annotation must have a nonnegative identifier indexing into this list.
iou_thresholds	The IOU (intersection over union) thresholds at which to compute the average precisions. This threshold determines whether a predicted bounding box and a ground truth bounding box are considered to match.

Member Function Documentation

add_row()

void turi::object_detection::average_precision_calculator::add_row	(	const std::vector< neural_net::image_annotation > &	predictions,
		const std::vector< neural_net::image_annotation > &	ground_truth
	)

Registers the predictions and ground truth annotations for one image.

evaluate()

variant_map_type turi::object_detection::average_precision_calculator::evaluate

(

)

Computes the average precision for each combination of class and requested IOU threshold.

Returns

A map of evaluation results keyed by metric.

The average precision can be interpreted as the area under the precision-recall curve.

average_precision_50 is a dictionary mapping class label to the average precision for that class label at 50% IOU.

average_precision is a dictionary mapping class label to the average precision for that class label, average across IOU thresholds from 50% to 95%.

mean_average_precision_50 is the mean across class labels of the average_precision_50 values.

mean_average_precision is the mean across class labels of the average_precision values.

---

The documentation for this class was generated from the following file:

• toolkits/object_detection/od_evaluation.hpp