Master AnnotatorMaster Annotator

Labeling Modes

Master Annotator provides four labeling modes, each optimized for a different annotation workflow. Choose the right mode based on the structure of your data and the level of precision you need.

Overview

When annotating trajectory data, different segments of motion have different characteristics. Some transitions require frame-by-frame precision, while long repetitive segments can be labeled in bulk. The four labeling modes let you match your annotation speed to the complexity of the data.

You can switch between modes at any time during annotation without losing any existing labels. Each mode affects only how new labels are applied — it never modifies labels you have already placed.

Single Mode

Single mode is the most precise labeling mode. Clicking a label (or pressing its keyboard shortcut) toggles that label on the current frame. The viewer stays on the same frame so you can review or change your decision before moving on.

How it works:

  1. Navigate to the frame you want to label.
  2. Click or press the shortcut for the desired label.
  3. The label is applied to the current frame. Pressing the same label again removes it (toggle behavior).
  4. Navigate manually to the next frame you want to annotate.

When to use Single mode

Single mode is ideal for transition frames where you need to inspect each frame carefully before deciding on the correct label.

Auto-Advance Mode

Auto-Advance mode applies a label and then automatically jumps forward by the configured step size. This creates a fluid annotation rhythm where you repeatedly press a shortcut key to label sequential frames without manually navigating.

How it works:

  1. Navigate to the starting frame.
  2. Click or press the shortcut for the desired label.
  3. The label is applied to the current frame and the viewer advances forward by the step size (e.g., 1, 5, or 10 frames).
  4. Continue pressing the same or a different label shortcut to annotate the next frame in sequence.

The step size is controlled by the Annotation Settings panel. Setting a step size of 1 labels every frame; larger values let you sample at a lower resolution for faster passes through long trajectories.

Range Mode

Range mode lets you select a start frame and an end frame, then paint a single label across every frame in that range. This is the fastest way to label long, uniform segments of motion.

How it works:

  1. Navigate to the first frame of the segment and click Set Start (or press the range start shortcut).
  2. Navigate to the last frame of the segment and click Set End (or press the range end shortcut).
  3. Select the label you want to apply. Every frame from start to end (inclusive) receives that label.

Range replaces existing labels

When you apply a label in range mode, it overwrites any previously applied labels on those frames. Review the start and end frames carefully before applying.

Keyframe Mode

Keyframe mode uses anchor points at key moments in the trajectory. You place labels on specific frames and the system interpolates labels across the gaps between keyframes. This works well for data where behavior changes happen at clear, identifiable moments.

How it works:

  1. Navigate to a frame where a behavior clearly starts and apply a label. This creates a keyframe anchor.
  2. Navigate to another frame where the behavior changes and apply a different label (another keyframe anchor).
  3. The system fills in the frames between your anchors — each frame inherits the label of the nearest preceding keyframe.

Efficient for structured tasks

If your trajectory follows a known task sequence (e.g., reach, grasp, lift, place), keyframe mode lets you mark just the transition points and have the rest filled automatically.

Mode Comparison

ModeSpeedPrecisionBest for
SingleSlowHighestTransition frames, corrections, ambiguous segments
Auto-AdvanceMediumHighSequential annotation, steady-pace labeling
RangeFastMediumLong uniform segments (e.g., "idle" or "moving")
KeyframeFastMediumStructured task sequences with clear transitions

Switching Modes

You can switch between labeling modes at any time using the mode selector in the annotation toolbar. Switching modes does not affect any labels you have already placed.

A common workflow is to use Range or Keyframe mode for an initial rough pass, then switch to Single mode to clean up transition boundaries.

Workflow Examples

Here are practical examples of how to combine labeling modes for common annotation tasks:

Example: Labeling a pick-and-place trajectory (500 frames)

  1. Rough pass with Keyframe mode — Scrub through the trajectory and identify the key transition points: the robot starts reaching at frame 0, grasps at frame 120, lifts at frame 180, moves at frame 250, places at frame 400, and releases at frame 450. Place a keyframe label at each transition. In under a minute, all 500 frames have a preliminary label.
  2. Refine boundaries with Single mode — Switch to Single mode and navigate to each transition zone (e.g., frames 115-125 around the reach-to-grasp boundary). Step frame-by-frame to find the exact frame where the behavior changes and correct any mislabeled frames.
  3. Quick verification with playback — Press Space to play back the trajectory. Watch the 3D viewer and timeline colors to spot any gaps or incorrect segments. Jump to any issues and fix them in Single mode.

Example: Labeling repetitive actions (2000+ frames)

  1. Identify segments with Range mode — Scrub through to identify long uniform sections (e.g., frames 0-300 are all "idle"). Set the start and end frame, apply the label — 300 frames labeled in three clicks.
  2. Fill remaining gaps with Auto-Advance — Set step size to 1 and switch to Auto-Advance mode. Navigate to an unlabeled section and press the label shortcut repeatedly. Each press labels a frame and advances to the next — a rhythmic workflow for sequences that change label every few frames.
Column Mapping3D Viewer