Release Notes | Create a cuboid from a 2D shape
Updated at January 9th, 2024
Break the 3D wall 🧱
With this release, you can now create a 3D cuboid from within the 2D workspace, projected based on the camera and location data in 3D space.
Feature released October 24th, 2023
A new option has been added to the Context (right-click) menu. In the menu, a new option has been added, named “Create 3D Cuboid”. This option will create a new 3D cuboid at the same point in space as occupied by the 2D shape.
How it works
To improve efficiency while annotating Sensor Fusion workloads, this new function streamlines the shape creation process between 2D bounding boxes and 3D Cuboids. As you begin annotating, rather than needing to first create the cuboid, then cross-reference its position in 2D space, you can begin by placing your bounding boxes, and using the “Create 3D Cuboid” feature to automatically place a cuboid in 3D space.
In the image above, the Cuboid created based on the white car is misaligned with the actual shape. While the cuboid will be created on the same plane as the camera, the positioning can be off due to perspective related to a “flat” plane. This is easily corrected by using your movement hotkeys to reposition the cuboid as it's created.Â
Key uses
In Sensor Fusion workloads, this feature enables a powerful use case. When comparing objects visible in 2D versus 3D, cameras will generally show objects from further away when compared to LIDAR or RADAR. This means that you can create shapes in 2D even when they aren't visible in 3D on the current frame, and the cuboids will appear in place on that shape.
In some cases, overlapping bounding boxes will prevent the cuboid from being placed in the correct spot. In these scenarios, the best bet is to enable either Accumulated Frames, or advance the frames until the 3D sensor detects the object.
See it in action
Click here for a detailed tutorial
Â