CARLA-GENERATOR

This is a co-simulator for autonomous driving that can be used to generate KITTI-compatible and semantic3D-compatible datasets. If you want to realize real-time network simulation, please cooperate with our sub-module NS3-DSRC, NS3-NR-C-V2X.

If you are interested in error concealment, please visit our LiDAR Point Cloud Error Concealment System.

Installation

Requirements

All the codes are tested in the following environment:

OS (Ubuntu 20.04 or Windows 10)
CARLA 0.9.12
Python 3.7.13+

Quick demo

a. Build environment

conda create -n carla-generator python==3.7.13 -y
conda activate carla-generator

b. Clone repository

git clone https://github.com/S-kewen/carla-generator
cd carla-generator
pip install -r requirements.txt

c. Install CARLA

We recommend using the binary release, you can also building CARLA from source code.

Windows 10

curl -0 https://carla-releases.s3.eu-west-3.amazonaws.com/Windows/CARLA_0.9.12.zip --output CARLA_0.9.12.zip
tar -xf CARLA_0.9.12.zip
pip3 install WindowsNoEditor\PythonAPI\carla\dist\carla-0.9.12-cp37-cp37m-win_amd64.whl
WindowsNoEditor\CarlaUE4.exe

Ubuntu 20.04

wget https://carla-releases.s3.eu-west-3.amazonaws.com/Linux/CARLA_0.9.12.tar.gz
tar -zxvf CARLA_0.9.12.tar.gz
cd CARLA_0.9.12
pip3 install CARLA_0.9.12/PythonAPI/carla/dist/carla-0.9.12-py3.7-linux-x86_64.egg
./CARLA_0.9.12/CarlaUE4.sh

d. Run

python generator.py --s NS3.ENABLE False # simple demo
python generator.py --s NS3.ENABLE True # with NS-3 simulator

For specific configuration, please refer to config.yaml.

Offline simulation

You also can store all the data and then run the NS-3 sub-module for network simulation.

python exp_offline_ns3.py --zmqPort {your_zmq_port} --path {your_dataset_path} --config {your_config_file}

Moreover, we support network simulation for the KITTI odometry dataset.

python exp_offline_ns3_kitti.py --zmqPort {your_zmq_port} --path {your_dataset_path} --config {your_config_file}

Output structures

├── save_directory_name
│   ├── ImageSets
│   │   ├── test.txt
│   │   ├── train.txt
│   │   ├── trainval.txt
│   │   ├── val.txt
│   ├── object
│   │   │   ├── training
│   │   │   │   ├── calib
│   │   │   │   ├── carla_label
│   │   │   │   ├── image_2
│   │   │   │   ├── image_label_2
│   │   │   │   ├── label_2
│   │   │   │   ├── location
│   │   │   │   ├── ns3
│   │   │   │   ├── packet
│   │   │   │   ├── planes
│   │   │   │   ├── ply
│   │   │   │   ├── semantic3d_label
│   │   │   │   ├── semantic3d_xyzirgb
│   │   │   │   ├── velodyne
│   │   │   │   ├── velodyne_compression
│   │   │   │   ├── velodyne_fg
│   │   │   │   ├── velodyne_ground_removal
│   │   │   ├── CAR2
│   │   │   │   ├── ...
│   │   │   ├── ...
│   ├── config.yaml

Limitations

We are happy to improve this project together, please submit your pull request if you fixed these limitations.

Calib: Our sensors are installed in a fixed location and can not provide calibration replacement.
LiDAR Intensity: The CARLA LiDAR sensor only provides virtual intensity without considering the material.

Acknowledgment

Part of our code refers to the work DataGenerator.

Contribution

welcome to contribute to this repo, please feel free to contact us with any potential contributions.