🚗 CARLA Dataset Tools

📦 A comprehensive data collection and labeling toolkit for CARLA Simulator

Data collection and labeling tools for CARLA Simulator. This toolkit provides an efficient pipeline for generating high-quality autonomous driving datasets with support for multiple sensor types and standard dataset formats.

⚠️ Important: All raw data generated by this tool uses a right-hand coordinate system.

🌟 This work is part of the CarlaFLCAV project. Feel free to give us a star!

✨ Key Features

✅ Multi-Sensor Support: RGB Camera, Semantic Segmentation, LiDAR, Semantic LiDAR, Radar, IMU, GNSS
✅ V2X Communication: Vehicle-to-Everything (V2X) sensor support for V2V and V2I scenarios
- V2X CAM sensors (ETSI standard) for automatic message generation
- V2X Custom sensors for manual message broadcasting
- Realistic wireless channel modeling with path loss simulation
✅ Infrastructure Support: Roadside unit (RSU) simulation for V2X scenarios
- Static infrastructure actors with sensor attachment
- One-way message broadcasting from infrastructure to vehicles
- Support for V2I (Vehicle-to-Infrastructure) communication
✅ Multiple Dataset Formats: KITTI Object, YOLOv5, Argoverse
✅ Flexible Configuration: YAML-based configuration with validation
✅ Route Editor: Interactive tool for creating custom vehicle routes with topology-aware path planning
✅ BEV Map Capture: Tool for generating bird's-eye view maps of entire CARLA maps
✅ Synchronized Recording: Synchronized multi-vehicle and multi-sensor data collection
✅ Visualization Tools: Built-in point cloud and data visualization utilities
✅ Autopilot Integration: Automatic vehicle control using CARLA's traffic manager

🚀 30-Second Quick Start

bash
# 1. Clone and install
git clone https://github.com/KevinLADLee/carla_dataset_tools.git
cd carla_dataset_tools
pip3 install -r requirements.txt

# 2. Set environment variable
export CARLA_ROOT=/path/to/your/carla

# 3. Start CARLA
cd $CARLA_ROOT && ./CarlaUE4.sh

# 4. Record data (in a new terminal)
cd carla_dataset_tools
python3 data_recorder.py --profile kitti

# 5. Generate labels 
python3 label_tools/kitti_objects_label.py -r record_YYYY_MMDD_HHMM -v vehicle_1st

# 6. Visulize output dataset
python3 tools/viz_lidar.py --type kitti --source dataset/record_YYYY_MMDD_HHMM/vehicle_1st/kitti_object/training/velodyne

📚 Documentation

For Users

User Guide - Complete installation and usage guide

Installation and prerequisites
Recording data with different profiles
Route Editor - Create custom vehicle routes interactively
Generating labels (KITTI, YOLOv5, Argoverse)
Data visualization
Troubleshooting

中文用户指南 - 完整的安装和使用指南

For Developers

Developer Guide - Technical documentation for developers

Architecture overview
Configuration system details (maps, weather, sensors)
API reference
Extending the toolkit
Development workflow

中文开发者指南 - 开发者技术文档

⚙️ Configuration Profiles

Pre-configured profiles for different dataset styles:

default - General purpose with multiple vehicles and sensors
kitti - KITTI-style (Velodyne HDL-64E, standard cameras)
argoverse - Argoverse-style with ring cameras
simple - Minimal configuration for testing
route_example - Demonstrates vehicle route following feature
v2x_test - V2X communication testing (V2V and V2I scenarios)
bev - Bird's-eye view camera configuration for map capture

bash
# List available profiles
python3 tools/config_list.py

# Use a profile
python3 data_recorder.py --profile kitti

# Create custom vehicle routes
python3 tools/editor_route.py --map Town02 --name my_route

# Capture BEV map of entire CARLA map
python3 tools/capture_map_bev.py --map Town02 --output ./bev_output

# Validate a configuration
python3 tools/config_validate.py --profile kitti

See Developer Guide for complete configuration reference.

📊 Supported Dataset Formats

KITTI Object Detection - 3D bounding boxes with calibration
YOLOv5 - 2D bounding box annotations
Argoverse - Ring camera setup (experimental)

🔧 Prerequisites

CARLA Simulator >= 0.9.16
Python >= 3.8
Operating System: Linux (recommended) / Windows

Download CARLA: https://github.com/carla-simulator/carla/releases

📁 Project Structure

carla_dataset_tools/
├── config/                      # Configuration management
│   ├── config_manager.py        # YAML config loader and validator
│   └── profiles/                # Pre-configured profiles (default, kitti, argoverse, simple, route_example)
├── docs/                        # Documentation
│   ├── USER_GUIDE.md            # User guide (English)
│   ├── USER_GUIDE_CN.md         # User guide (Chinese)
│   ├── DEVELOPER.md             # Developer guide (English)
│   └── DEVELOPER_CN.md          # Developer guide (Chinese)
├── label_tools/                 # Labeling scripts (KITTI, YOLO, Argoverse)
├── recorder/                    # Core recording modules
├── routes/                      # Vehicle route definitions (YAML + PKL)
├── core/                        # Core shared modules (geometry, transform, logger)
├── tools/                       # CLI utility scripts
│   ├── editor_route.py          # Route editor
│   ├── capture_map_bev.py       # BEV map capture tool
│   ├── viz_lidar.py             # LiDAR visualization
│   └── ...                      # Other utilities
└── data_recorder.py             # Main recording script

🤝 Contributing

Contributions are welcome! Areas for contribution:

Additional dataset format support (nuScenes, Waymo, etc.)
Enhanced documentation and examples
Bug fixes and performance improvements
New sensor types or features

Please submit pull requests to the main repository.

📖 Citation

If you use this tool in your research, please cite:

bibtex
@article{wang2022federated,
  title={Federated deep learning meets autonomous vehicle perception: Design and verification},
  author={Wang, Shuai and Li, Chengyang and Ng, Derrick Wing Kwan and Eldar, Yonina C and Poor, H Vincent and Hao, Qi and Xu, Chengzhong},
  journal={IEEE network},
  volume={37},
  number={3},
  pages={16--25},
  year={2022},
  publisher={IEEE}
}