A semi-automatic, web-based toolbox for annotating 3D bounding boxes in full-surround, multi-modal sensor data streams.
3D Bounding Box Annotation Toolbox (3D-BAT) is an open-source tool for creating 3D bounding box annotations on point cloud and image data. It solves the problem of efficiently labeling multi-sensor, full-surround data streams for applications like autonomous vehicle perception and robotics. The tool provides a semi-automatic, web-based interface to streamline the annotation workflow.
Researchers and engineers in autonomous driving, robotics, and computer vision who need to create high-quality 3D labeled datasets from LiDAR and camera data.
Developers choose 3D-BAT for its comprehensive feature set supporting full-surround data, AI-assisted labeling, and active learning integration. Its web-based, customizable platform offers a flexible alternative to proprietary annotation tools.
3D Bounding Box Annotation Tool (3D-BAT) Point cloud and Image Labeling
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Integrates machine learning for AI-assisted labeling and active learning support, significantly reducing manual annotation time as highlighted in the README's features and news sections.
Supports synchronized labeling of 3D point clouds and 2D images with features like 3D-to-2D projection, essential for autonomous driving datasets.
Designed for full-surround data streams with side views and navigation tools, enabling comprehensive object labeling from all angles around sensors.
Configurable for custom datasets, object classes, and attributes, allowing adaptation to specific research needs via the custom data annotation tutorial.
Requires installation of npm, conda, Python packages, and multiple commands, which can be complex and time-consuming for non-expert users.
The README notes that comprehensive documentation is 'available soon', leaving users to rely on limited tutorials and source code for guidance.
Heavily optimized for autonomous driving and V2X data; may not seamlessly support other domains like indoor robotics without significant customization.