Thresholds Statistics

The Thresholds Statistics view is where you configure how confident a model needs to be before its predictions are presented as suggestions during annotation. Each label can have its own threshold and default value.

Why thresholds matter

When a model predicts a label, it also reports how confident it is — a number between 0 and 1. A prediction with confidence 0.95 means the model is fairly certain. A prediction with confidence 0.4 means it is guessing.

The confidence threshold you set here determines the boundary. Predictions at or above the threshold are shown as suggestions in the annotation view. Predictions below it are not surfaced — the annotator sees the default value instead.

Tip

Setting the right threshold is a judgment call. A high threshold means fewer suggestions but more reliable ones. A low threshold means more suggestions but more noise. Start conservative and adjust as you learn to trust the model.

Selecting a configuration

At the top of the page, a dropdown labeled Select Config lets you choose which annotation configuration to edit. Each configuration corresponds to a specific image position in your inspection setup, identified by its image label or image ID.

Note

Annotation configurations are tied to image IDs. Different cameras or image positions on a rig may have different labels and thresholds because they capture different aspects of the product.

Editing thresholds and defaults

Once you select a configuration, the page shows each annotation label within it.

For each label you can adjust:

Setting	Description
Label name	The human-readable name of this label, as defined in the label map.
Default value	The value pre-selected during annotation when no prediction exists or when the prediction is below the threshold. Choose from the available options for this label.
Confidence threshold	A number between 0 and 1. Predictions with confidence at or above this value are shown as suggestions. Enter it directly or use the slider.

Caution

If a label name appears in red with “no label found”, the label map is missing an entry for that label ID. Contact your administrator to update the label configuration.

Saving changes

1. Select the configuration you want to edit from the dropdown.
2. Adjust default values and thresholds for each label as needed.
3. Click Save at the bottom of the page.
4. Confirm the result. A green notification appears if the save succeeded, or a red one if something went wrong.

Changes take effect immediately for all users. The next time someone opens the annotation view, they will see the updated thresholds and defaults.

What’s next?

Understand model behavior Use confidence scores and filters to evaluate a model.

Compare model versions Check whether a new model is better than the current one.

Glossary Definitions for terms used in EyVz.