Label Management

Labels are the vocabulary of your annotation task. Each label represents a distinct behavior or state in your robot trajectory. Master Annotator gives you full control over label names, colors, keyboard shortcuts, and export values.

Overview

Labels are defined at the project level and shared across all datasets in that project. This ensures consistent annotation across your entire dataset collection. You manage labels from the Labels tab in your project settings.

Each label has four properties:

• Name — A human-readable label shown in the UI (e.g., "Reaching", "Grasping", "Idle").
• Color — A visual color used in the timeline, 3D viewer highlights, and label buttons.
• Keyboard shortcut — A key binding for fast annotation without mouse clicks.
• Export value — The string written to the CSV when exporting labeled data.

Creating Labels

To create a new label, open your project's Labels tab and click the Add Label button. Fill in the label name, select a color, and optionally assign a keyboard shortcut and export value.

Label names must be unique within a project. Keep them short and descriptive — they appear on buttons in the labeling toolbar and need to be readable at a glance.

Label Colors

Master Annotator provides 10 color options for labels. Colors serve as the primary visual identifier in the annotation interface:

• Timeline segments — Labeled frames appear as colored bands on the timeline scrubber.
• Label buttons — Each button in the labeling toolbar displays its assigned color.
• Frame indicator — The current frame's label color is shown prominently in the interface.

Keyboard Shortcuts

Assigning keyboard shortcuts to labels dramatically speeds up annotation. Instead of clicking buttons, you press a single key to apply a label. Two types of shortcuts are available:

• Positional (1–9) — Labels are automatically assigned shortcuts based on their position in the list. The first label gets 1, the second gets 2, and so on up to 9.
• Custom (A–Z) — You can assign a specific letter key to any label. For example, assign G to "Grasping" for a mnemonic shortcut.

Custom letter shortcuts take precedence over positional shortcuts. If you assign R to a label, pressing R applies that label rather than triggering Range mode.

Export Values

Each label has an export value — the string that appears in the exported CSV file. By default, the export value matches the label name, but you can set it to any custom string.

This is useful when your downstream pipeline expects specific values:

• Numeric codes: set export value to 0, 1, 2 instead of label names.
• Abbreviated values: use REACH instead of "Reaching toward object".
• Pipeline-specific keys: match whatever format your training scripts expect.

See CSV Export for details on how export values are used in the output file.

Reordering Labels

Labels can be reordered by dragging and dropping them in the Labels tab. The order determines:

• Button order — Labels appear in the labeling toolbar in the same order as the list.
• Positional shortcuts — The first label in the list gets shortcut 1, the second gets 2, and so on.
• Export column order — When exporting with per-label columns, the column order matches the label order.

Reordering labels does not affect any existing annotations — only the display order and positional shortcuts change.

Usage Tracking

The Labels tab shows usage information for each label, indicating which datasets in the project have annotations using that label. This helps you understand:

• Which labels are actively in use across your datasets.
• Whether any labels have never been applied (potential candidates for removal).
• The distribution of annotations across your label vocabulary.

Editing and Deleting

You can edit any label property (name, color, shortcut, export value) at any time. Edits take effect immediately in the labeling interface.

Deleting a label is a destructive action:

• The label is removed from the project's label schema.
• Any frames annotated with that label across all datasets lose their label assignment.
• This action cannot be undone.

Best Practices

• Define labels before annotating — Set up your complete label schema before starting annotation to avoid mid-workflow changes.
• Use mnemonic shortcuts — Assign letter shortcuts that match the label name (e.g., G for Grasp, L for Lift) to build muscle memory.
• Keep the label set small — Fewer, well-defined labels lead to more consistent annotations than many overlapping categories.
• Set export values upfront — Define export values before annotating so the exported data matches your pipeline requirements without post-processing.
• Use templates — If you reuse the same label schema across projects, save it as a label template.