Appen's 3D Point Cloud Annotation tool provides the perfect balance of highly technical annotation capabilities and a simple, easy to use annotator interface that allows for fast and high-quality annotations.
Important Note: This new tool is part of our managed service offerings. If you are interested in learning more about the Point Cloud Annotation tool, please contact your Appen Customer Success Manager or Account Executive.
Note: For Dedicated customers, this feature is currently not available On-Premises and only available via the cloud multi-tenant Appen Data Annotation Platform.
How it Works
The Point Cloud Annotation tool fuses any 3D sensor data with 2D camera images to build an intuitive annotation interface. Contributors can then create 3D annotations to bring to your models across a diverse set of industries.
Figure 1: Appen Point Cloud Annotation Tool via Preview Page
Point Cloud Annotation Features
- Draw and Track Cuboids
- Annotators can draw and track cuboids on objects in sequences of point cloud data featuring sensor-fused 2D images for annotators to better visualize their annotations.
- LiDAR-Assisted 2D Bounding Box Annotation
- Create bounding boxes on images with the assistance of Lidar (does not require association to cuboids in Lidar).
- LiDAR Semantic Segmentation
- The tool allows the assigning of a class and an instance ID for each 3D point.
- Ability to Validate and Improve Model Predictions
- If you have pre-labeled 3D annotations, you can bring these to the tool for human annotators to review. This will also help to accelerate data annotation and get precise metrics on your model’s performance.
- Track Occlusion and Truncation
- The tool allows for levels on each object for each frame for more comprehensive detection of the object.
- Additionally, the tool supports a customized attribute list that can contain occlusion and truncation, but also other attributes as desired
- Enable Faster Data Annotation with Machine Assistance
- The tool offers enhanced object tracking using interpolation and one-click box cuboid auto-adjustment with machine learning generated clustering.
- Leverage Industry-leading Interaction Design
- Our design allows annotators to quickly navigate the scene, understand the context and efficiently label the data. Interactive multi-angle views and sensor fusion allow annotators to choose the best view for each object.
Input Data (Source Data)
The Point Cloud Annotation tool accepts various open source data formats; please see below for examples:
If you have a specific data format that you’d like to use with the Point Cloud Annotation tool, please reach out to your Customer Success Manager or Account Executive for more information.
Output Data (Results)
The output annotation result is a link to a hosted file.
For more information on output details, please reach out to your Appen Customer Success Manager or Account Executive.