ELocatorMetrics Class

Collection of metrics used to evaluate the results of a ELocator tool.

Namespace: Euresys::Open_eVision_2_15::EasyDeepLearning

Methods

Constructs an ELocatorMetrics object.
Average Precision when matching prediction and ground truth with a minimum IoU of ELocator::SameLabelMaxOverlap.
Average precision when matching prediction and ground truth with a minimum IoU of 0.5.
Best weighted F-Score achievable by changing the detection threshold.
Best weighted F-Score achievable with the corresponding threshold.
Detection threshold that gives the best weighted F-Score (ELocatorMetrics).
Best weighted precision achievable by changing the detection threshold.
Best weighted precision achievable with the corresponding threshold.
Detection threshold that gives the best weighted precision (ELocatorMetrics).
Best weighted recall achievable by changing the detection threshold.
Best weighted recall achievable with the corresponding threshold.
Detection threshold that gives the best weighted recall (ELocatorMetrics).
Threshold for detected objects.
Error of the EasyLocate training algorithm (only available for the training metrics, see EDeepLearningTool).
F-Score (harmonic mean of ELocatorMetrics::Recall and ELocatorMetrics::Precision).
Image accuracy: the proportion of images that are correctly detected to contain or not objects, regardless of their labels.
Average of the label intersection over union.
Label.
Average precision for detection of objects from the given label.
F-Score for the given label.
Label intersection over union (IoU). This is the average intersection over union between predicted and ground truth objects of the given label.
Precision for the given label.
Recall for the given label.
Number of images with objects that are badly predicted as containing no object. This is the number of false negatives at the image level.
Number of images without objects that are badly predicted as containing objects. This is the number of false positives a the image level.
Number of correctly detected objects.
A correctly detected object is a predicted object that has an IoU bigger than 0.5 with a ground truth object of the same label.
A label can be specified to obtain the number of correctly detected objects for that label. Otherwise, the number of correctly detected objects is for all the labels.
Number of images containing objects that are correctly predicted as containing objects, regardless of the labels of the objects. This is the number of true positives at the image level.
Number of images without objects that are correctly predicted as containing no object. This is the number of true negatives at the image level.
Number of detected objects.
A label can be specified to obtain the number of detected objects for that label. Otherwise, the number of detected objects is for all the labels.
Number of labels recognized by the ELocator tool that produced these metrics.
Number of undetected objects.
An undetected object is a ground truth object that is not matched to any predicted object of the same label with an IoU bigger than 0.5.
A label can be specified to obtain the number of undetected objects for that label. Otherwise, the number of undetected objects is for all the labels.
Precision (proportion of detected objects that are correct).
Recall (true positive rate, proportion of ground truth object that are correctly detected).
Weighted average of the F-Score for each label.
Weighted average of the precision (proportion of detected objects that are correct) for each label.
Weighted average of the recall (true positive rate) for each label.
Indicates whether the object contains at least one result.
Loads a locator metric. The given ESerializer must have been created for reading.
Assignment operator.
Saves a locator metric. The given ESerializer must have been created for writing.
Serializes the metrics.
Threshold for detected objects.

ELocatorMetrics Class

Collection of metrics used to evaluate the results of a ELocator tool.

Namespace: Euresys.Open_eVision_2_15.EasyDeepLearning

Properties

Average Precision when matching prediction and ground truth with a minimum IoU of ELocator::SameLabelMaxOverlap.
Average precision when matching prediction and ground truth with a minimum IoU of 0.5.
Threshold for detected objects.
Error of the EasyLocate training algorithm (only available for the training metrics, see EDeepLearningTool).
F-Score (harmonic mean of ELocatorMetrics::Recall and ELocatorMetrics::Precision).
Image accuracy: the proportion of images that are correctly detected to contain or not objects, regardless of their labels.
Average of the label intersection over union.
Number of images with objects that are badly predicted as containing no object. This is the number of false negatives at the image level.
Number of images without objects that are badly predicted as containing objects. This is the number of false positives a the image level.
Number of images containing objects that are correctly predicted as containing objects, regardless of the labels of the objects. This is the number of true positives at the image level.
Number of images without objects that are correctly predicted as containing no object. This is the number of true negatives at the image level.
Number of labels recognized by the ELocator tool that produced these metrics.
Precision (proportion of detected objects that are correct).
Recall (true positive rate, proportion of ground truth object that are correctly detected).

Methods

Constructs an ELocatorMetrics object.
Best weighted F-Score achievable by changing the detection threshold.
Best weighted F-Score achievable with the corresponding threshold.
Detection threshold that gives the best weighted F-Score (ELocatorMetrics).
Best weighted precision achievable by changing the detection threshold.
Best weighted precision achievable with the corresponding threshold.
Detection threshold that gives the best weighted precision (ELocatorMetrics).
Best weighted recall achievable by changing the detection threshold.
Best weighted recall achievable with the corresponding threshold.
Detection threshold that gives the best weighted recall (ELocatorMetrics).
Label.
Average precision for detection of objects from the given label.
F-Score for the given label.
Label intersection over union (IoU). This is the average intersection over union between predicted and ground truth objects of the given label.
Precision for the given label.
Recall for the given label.
Number of correctly detected objects.
A correctly detected object is a predicted object that has an IoU bigger than 0.5 with a ground truth object of the same label.
A label can be specified to obtain the number of correctly detected objects for that label. Otherwise, the number of correctly detected objects is for all the labels.
Number of detected objects.
A label can be specified to obtain the number of detected objects for that label. Otherwise, the number of detected objects is for all the labels.
Number of undetected objects.
An undetected object is a ground truth object that is not matched to any predicted object of the same label with an IoU bigger than 0.5.
A label can be specified to obtain the number of undetected objects for that label. Otherwise, the number of undetected objects is for all the labels.
Weighted average of the F-Score for each label.
Weighted average of the precision (proportion of detected objects that are correct) for each label.
Weighted average of the recall (true positive rate) for each label.
Indicates whether the object contains at least one result.
Loads a locator metric. The given ESerializer must have been created for reading.
Assignment operator.
Saves a locator metric. The given ESerializer must have been created for writing.
Serializes the metrics.