3D Cuboid Tool

Documentation Index

Fetch the complete documentation index at: https://avala.ai/docs/llms.txt

Use this file to discover all available pages before exploring further.

The 3D Cuboid tool draws three-dimensional bounding boxes in the point cloud viewer. It is the primary tool for annotating objects in LiDAR point cloud data, commonly used in autonomous driving and robotics.

​

When to Use

Use 3D cuboids when:

• You are annotating LiDAR point cloud data
• The task requires 3D object detection (e.g., vehicles, pedestrians, cyclists)
• You need position, dimensions, and orientation (heading) for each object
• The downstream model expects 3D bounding box labels

Consider a different tool when:

• You are working with 2D images — use the Bounding Box tool
• You need lane or boundary annotations in 3D — use the Polyline tool
• You need semantic labels for the entire scene — use the Segmentation tool

​

Usage

1. Press C or select the Cuboid tool from the toolbar
2. Click in the point cloud to place the cuboid center
3. A default cuboid appears at the clicked location
4. Adjust the cuboid:
   • Drag corners: Resize length, width, and height
   • Drag edges: Adjust individual dimensions
   • Drag center handle: Reposition the cuboid
   • Drag rotation handle: Rotate around the Z-axis (yaw)
5. Assign a label

​

Adjusting Dimensions

​

Multi-Camera Verification

When camera data is available, 3D cuboids automatically project onto camera images:

1. Draw or adjust the cuboid in the 3D view
2. Check the camera panels to verify the projection aligns with the object
3. Adjust if the 2D projection does not match

This is one of the most effective quality checks for 3D annotations — the camera projection reveals errors in depth, height, and heading that are hard to see in the top-down view alone.

​

Shortcuts

​

Common Mistakes

• Box too large: The cuboid should tightly enclose just the object’s points, not a cloud of surrounding points
• Wrong ground alignment: The bottom face should sit on the ground plane — objects that float or sink below the ground indicate incorrect height
• Incorrect heading: The cuboid’s forward direction should match the object’s direction of travel, not be arbitrary
• Ignoring camera verification: Always check the camera projection when cameras are available

​

Advanced Tips

• Fit to points: Size the cuboid to tightly enclose the object’s points only
• Ground contact: Align the bottom face with the ground plane
• Orientation: Match the cuboid heading to the object’s direction of travel
• Use camera views to verify depth and height, which are harder to judge in the top-down 3D view
• Press R to reset the 3D camera view if you lose orientation
• For sequences, cuboids can be tracked across frames with interpolation between key frames

​

Related Tools

• Bounding Box — For 2D rectangular annotations on images
• Polyline — For lane markings and boundaries in 3D
• Keypoint — For pose estimation on 2D images
• Segmentation — For pixel-level semantic labels