3D Point Cloud Annotation

This tutorial covers annotating 3D point cloud data with cuboids and using multi-camera projection.

What You’ll Learn

• How to upload and view point cloud data
• How to draw 3D cuboids in point clouds
• How to use multi-camera views for verification
• How to track objects across frames

Prerequisites

• Point cloud data (MCAP, PCD, or PLY format)
• Camera images (optional, for multi-view)
• Camera calibration data (for projection)

Step 1: Upload Point Cloud Data

MCAP Files (Recommended)

1. Create a new dataset with type LiDAR
2. Upload your MCAP file containing:
   • Point cloud messages (e.g., sensor_msgs/PointCloud2)
   • Camera images (optional)
   • Calibration data (optional)

Individual Point Clouds

1. Create a dataset with type LiDAR
2. Upload PCD or PLY files
3. Files are treated as individual scans

Step 2: Open the 3D Viewer

1. Navigate to your dataset → Sequences
2. Click on a sequence to open the viewer
3. The interface shows:
   • 3D viewport: Point cloud visualization
   • Camera views: Synchronized camera images
   • Timeline: Frame navigation
   • Object list: Annotated objects

Step 3: Navigate the 3D View

Mouse Controls

Display Options

• Point size: Adjust point rendering size
• Color mode: Height, intensity, or RGB
• Ground plane: Toggle ground visualization

Step 4: Draw 3D Cuboids

1. Select the Cuboid tool from the toolbar
2. Click in the point cloud to place the cuboid center
3. Adjust the cuboid:
   • Drag corners: Resize dimensions
   • Drag edges: Adjust individual dimensions
   • Drag center: Move position
   • Drag rotation handle: Rotate around Z-axis
4. Assign a label from the dropdown
5. Click Save

Cuboid Tips

• Fit to points: Adjust cuboid to tightly enclose points
• Use camera views: Verify cuboid position in 2D
• Ground contact: Align bottom with ground plane
• Orientation: Match vehicle/object heading

Step 5: Multi-Camera Projection

When camera data is available:

1. 3D cuboids automatically project onto camera images
2. Verify projections in each camera view
3. Adjust cuboid position if projection is misaligned
4. Check occlusion and truncation

Calibration

If projections are misaligned:

1. Go to Settings → Camera Calibration
2. Adjust intrinsic/extrinsic parameters
3. Save calibration
4. Projections update automatically

Step 6: Object Tracking

For sequences with multiple frames:

1. Create a cuboid on the first frame
2. Navigate to the next frame
3. The object persists (tracking mode)
4. Adjust position for object movement
5. Repeat for each frame

Tracking Tips

• Interpolation: System can interpolate between keyframes
• ID consistency: Keep the same object ID across frames
• Lost objects: Mark as occluded when not visible

Step 7: Review and Export

1. Review annotations in the 3D viewer
2. Check multi-camera projections
3. Export in your preferred format:
   • KITTI: Standard AV format
   • nuScenes: Multi-frame format
   • Avala Format: Full metadata

Keyboard Shortcuts

Next Steps

• Configure camera calibration
• Learn about sequence workflows
• Export to different formats