3D-BAT (3D Bounding Box Annotation Toolbox) is a web-based, open-source tool designed for annotating point cloud and image data, primarily for autonomous driving and robotics research. It offers a comprehensive suite of features for efficient and accurate 3D object detection labeling, supporting multi-sensor data and AI-assisted workflows.

How It Works

3D-BAT utilizes a web-based architecture, allowing for platform-independent access. It supports full-surround annotations, including 3D bounding boxes, and offers features like AI-assisted labeling, interpolation, and 3D-to-2D label transfer. The tool is designed for extensibility, enabling custom datasets, classes, and attributes, with support for formats like OpenLABEL and V2X data.

Quick Start & Requirements

1. Install Node.js: Download from nodejs.org.
2. Clone Repository: git clone https://github.com/walzimmer/3d-bat.git && cd 3d-bat
3. Python Environment: conda create -n 3d-bat python==3.11.3, conda activate 3d-bat, pip install -r requirements.txt
4. Node.js Packages: npm install
5. Start Server: npm run start-server
6. Start Tool: npm run start
   • Prerequisites: Node.js (v10.15.0 recommended), Python 3.11.3, Conda.
   • Documentation: readthedocs (upcoming)

Highlighted Details

• AI-assisted labeling and active learning support.
• 3D to 2D label transfer and automatic tracking.
• Supports custom datasets, classes, attributes, and OpenLABEL format.
• Web-based, platform-independent, and extensible architecture.

Maintenance & Community

The project has seen recent updates in early 2024, including active learning support and V2X data labeling. It has been recognized with an IEEE Best Student Paper Award at ITSC'23. Further community engagement details are not explicitly provided in the README.

Licensing & Compatibility

The software is licensed under a permissive, non-commercial license from The Regents of the University of California. Commercial use requires explicit permission and contact with the UC San Diego Office of Innovation and Commercialization.

Limitations & Caveats

The license explicitly restricts commercial use without separate agreement. While extensible, the README notes that a comprehensive readthedocs documentation is still upcoming.