Velocity Graph
The Velocity Graph displays how an object's speed changes across a sequence, enabling users to detect annotation errors and physically implausible motion at a glance.
Use Cases
Autonomous Driving - Highway Vehicle Tracking When annotating fast-moving vehicles on a highway sequence, sudden spikes in the velocity graph immediately reveal frames where a cuboid was accidentally misplaced or where keyframe interpolation has overshot the realistic path - without needing to scrub through every frame manually.
Robotics - Moving Equipment In warehouse or manipulation datasets, the velocity graph helps reviewers confirm that robotic arms or moving equipment follow physically consistent motion paths, catching abrupt jumps that indicate incorrect keyframe placement across a sequence.
Agri-Tech - Monitoring Machinery Across Fields When tracking tractors or drones across long sequences, the velocity graph surfaces frames where the object's speed deviates unexpectedly, helping annotators distinguish genuine motion changes from annotation drift or copy-paste errors.
Aviation - Aircraft Annotation Quality Review When annotating aircraft sequences, the velocity graph gives auditors a rapid way to identify spikes in an object's speed profile and determine whether a sequence requires an additional round of internal QA - without scrubbing through every frame individually. In one instance, a delivery team reviewing flagged spikes discovered that the root cause was not an annotation error but LiDAR sensor jitter on specific frames - a finding that would have been difficult to isolate without the velocity graph surfacing the affected frames directly.
Benefits
For Annotators: Provides an immediate, at-a-glance view of an object's speed profile across the sequence without manually scrubbing through frames, reducing time per object during tracking and interpolation tasks.
For Customers (AI/ML & Perception Engineering): Enables direct visual validation of object motion data against expected real-world speed profiles. Higher confidence in motion data quality translates directly into more reliable velocity estimation and trajectory prediction model training.
Cost Efficiency: Surface-level motion anomalies that previously required frame-by-frame scrubbing to detect can now be identified in seconds from the graph, reducing QA time and the cost of late-stage corrections on motion-critical datasets.
Steps to Use
The Velocity Graph requires ego pose data and timestamped frames to compute velocity accurately. If either is absent, the Velocity Graph option will not appear in the right-click menu.
Create a cuboid annotation on the frame where the object appears most well defined.
Track the object across the sequence using keyframe interpolation, repositioning the cuboid every 5–7 frames as needed.
Once tracking is complete, select the cuboid and right-click to open the context menu.
Click Velocity Graph, a graph window appears
The graph plots velocity (m/s) on the Y-axis against absolute frame numbers on the X-axis. Forward velocity is displayed by default.
Hover over any point on the graph to reveal a crosshair and tooltip showing the exact frame number and velocity value. For occluded frame ranges where the cuboid was deleted between keyframes, the graph renders a gap in the line and the tooltip displays "Frame N · NA" with no velocity value.
To inspect dense sequences, scroll the mouse wheel over the graph to zoom. Click and drag to pan across the frame range after zooming in.
To return to the full sequence view, click the reset button to restore the default zoom and pan state. The zoom and pan state also resets automatically when a different cuboid is selected.
To explore velocity on a specific axis or plane, use the dropdown on the graph to select from Vx, Vy, Vz, or a combination. The Y-axis label updates to reflect the selection (e.g., "Velocity - Vxy (m/s)"). The name of the selected object appears in the graph heading.
Last updated