Metrics

The Metrics tab consists of a confusion matrix and a table view of the results.

Note: The Pattern tool does not have a Metrics tab.

Note: For the Segmenter tool, you can select which role you want to display in the confusion matrix and table.

Confusion Matrix

The confusion matrix is a table that compares the predicted values to actual applied labels. Each row in the matrix shows the number of features you labeled, which is the ground truth. Each column shows the number of features the tool has predicted.

Note: The confusion matrix and metrics calculations only include labeled images. The default confusion matrix displays data for all labeled images assigned a test role or unassigned role. Images used for training a tool must not be used when testing the accuracy of a tool. For this reason, the confusion matrix filters any views with that are assigned the train role.

The color scheme of the confusion matrix provides a quick way for you to see the distribution of the views in the matrix.

The color of a cell is the ratio of predicted features to the total number of labeled features per class. A darker color means a higher ratio and a lighter color means a lower ratio.

For example, the cell for features labeled as Dark Magic and predicted as Dark Magic contains the number 22. This cell is dark red because 22 is 100% of the total number of features labeled as Dark Magic. The cell for features labeled as French Roast and predicted as French Roast contains the number 4. This cell is a lighter red because 4 is only 66% of the 6 total features labeled as  French Roast.

A diagonal line of dark cells in the confusion matrix shows that the tools predictions are matching the expected labels.

You can use the confusion matrix to identify the input views that produced specific results by clicking on one or more cells. This filters the images in the image manager and applies a blue border around the column, indicating the filtered status.

Table View

The Table view provides detailed metrics about each feature class.


Column	Description
Class	Name of the feature class.
Predicted	Number of features the tool predicted as this class when processing all views.
Unclassified	Number of features the tool did not classify.
Labeled	Number of features in this feature class included in all views.
Trained	Number of features in this feature class included in the training set.
Recall	Percentage of labeled features or classes that the tool correctly identified.
Precision	Percentage of detected features or classes that match the labeled feature or class.
F-Score	Represents the overall accuracy of the tool. F-Score is the harmonic mean of the Recall and Precision scores, which provides a single metric that balances both Precision and Recall. The value of the F-score is between 0 to 1, with 1 being a better score.