Filtering

The filter pane sits on the left side of the Home view. It controls which images appear in the grid. No filter change takes effect until you click Apply filters.

Filter fields

The table below lists every standard filter field in the filter pane, from top to bottom.

Filter	What it does	Accepted values
From	Start of the time window.	Date and time. Defaults to midnight today (or yesterday if before 2:00 AM).
To	End of the time window.	Date and time. Leave empty for no upper limit.
Annotation	Match images by human annotation.	label:value format, e.g. defect:scratch. Multiple entries allowed.
Prediction	Match images by model prediction.	label:value format, e.g. class:ok. Multiple entries allowed.
Confidence (%)	Show only images whose prediction confidence falls within this range.	Min and max percentage (0–100). Leave empty for no limit.
Rig ID	Show images from specific inspection rigs.	One or more rig identifiers.
Camera ID	Show images from specific cameras.	One or more camera identifiers.
Image ID	Show a specific image sequence number.	One or more image identifiers.
Frame trigger	Match a frame trigger identifier.	One or more trigger identifiers.
Custom tags	Match user-defined tags.	One or more tag names.
Model ID	Filter by which model produced predictions.	Dropdown selection. All shows images from any model.
Annotated by	Filter by which user annotated the image.	Dropdown selection. All shows images annotated by anyone.

Site-specific fields

The filter pane may show additional filter fields configured for your specific installation. These appear above the standard filters and are tailored to your site’s data. They accept the same input patterns as the standard text filters, including multiple values and negation.

Note

Site-specific fields vary between deployments. You will recognise your own fields when you see them in the filter pane.

How values combine within a single filter

For most filters, entering multiple values in the same field means OR — an image matches if it satisfies any of the listed values.

Example: Rig ID = 1, 2, 3 shows images from rig 1 or rig 2 or rig 3.

The Prediction filter is different: multiple prediction entries combine with AND logic. An image must match all specified predictions.

Example: Prediction = defect:scratch, quality:reject shows only images where the model predicted both defect:scratch and quality:reject.

How different filters combine with each other

Filters across different fields always combine with AND logic. An image must satisfy every active filter simultaneously.

Example: Rig ID = 1 and Camera ID = 2 and Confidence ≥ 50% shows only images from rig 1, camera 2, with at least 50% confidence.

Negation (click to exclude)

After adding a value to a text filter, clicking on it toggles it to a negated state. Negated values are prefixed with ! and appear visually in red. A negated value means “exclude images matching this value.”

Example: Adding 41 to Rig ID and then clicking on it turns it into !41 — this excludes all images from rig 41.

When a filter contains both positive and negated values, the logic is: include images matching any of the positive values and exclude images matching any of the negated values.

Tip

Negation works on all text-based filters, including Annotation, Prediction, Rig ID, Camera ID, Image ID, Frame trigger, Custom tags, and site-specific fields.

Date range — From and To

Field	Default	Behaviour
From	Midnight of the current day. If it is before 2:00 AM, defaults to midnight of the previous day.	Sets the start of the time window.
To	Empty (no upper limit).	Sets the end of the time window. When empty, images up to the most recent are shown.

The timestamps refer to when the image was captured or ingested into the system.

Confidence range vs. Prediction threshold slider

These two controls look similar but behave very differently.

Control	Type	Effect
Confidence (%) min/max	Data filter	Images outside the range are removed from results entirely. You will not see them in the grid.
Only show prediction above threshold (%)	Display filter	Hides prediction labels below the threshold on image cards, but does not remove any images from the grid.

Caution

The prediction threshold slider is useful for reducing visual noise on image cards without losing access to any images. The confidence range filter permanently excludes images from the results.

Annotation and Prediction filter format

Both filters use the label:value format. Type a value and press Enter or Tab to add it.

Behaviour	Annotation filter	Prediction filter
Multiple positive entries	OR — image matches if any annotation matches	AND — image must match all specified predictions
Negated entries (prefixed with !)	Exclusion — images with the matching annotation are removed	Exclusion — images with the matching prediction are removed
Supported types	Classification annotations	Classification and object detection predictions

Example (Annotation): defect:scratch, defect:dent shows images annotated with either defect:scratch or defect:dent.

Example (Prediction): quality:ok, surface:clean shows images where the model predicted both quality:ok and surface:clean.

Tip

The dropdown suggests known labels as you type, making it easier to use the correct format.

Show/Hide toggles

Two radio-button groups let you filter by annotation and prediction status:

Toggle	Options	Effect
Show/Hide annotated images	All, Annotated, Unannotated	Show all images, only those with human annotations, or only those without.
Show/Hide predicted images	All, Predicted, Unpredicted	Show all images, only those with model predictions, or only those without.

Hide skipped annotations

When enabled (the default), images whose annotations were marked as “skipped” during annotation are hidden from results. Disable this if you want to review skipped images.

Auto-update

When enabled, Eydara polls the database every second for new images and refreshes the grid automatically. This is useful when actively monitoring incoming images during production.

Caution

Auto-update adds continuous database load. Disable it when you are not actively monitoring new images.

Apply Filters / Clear Filters

Filter changes are not applied until you click Apply filters. This lets you adjust multiple filters before running the query.

Clear filters resets all fields to their default values, including the default time range.

Tip

Apply filters appears both at the top and bottom of the filter pane for convenience.

Filter profiles

If filter profiles are configured for your site, you can save and load filter configurations. This lets you quickly switch between commonly used filter combinations without re-entering values each time.

What’s next?

Home Return to the main image browsing view.

How-To: Evaluate a model Use filters to investigate model predictions.