Co3SOP: A Synthetic Benchmark for Collaborative 3D Semantic Occupancy Prediction in V2X Autonomous Driving

Paper | GitHub

Co3SOP is a high-resolution synthetic benchmark designed for Collaborative 3D Semantic Occupancy Prediction in V2X-enabled autonomous driving.

While single-vehicle perception is often limited by occlusions, restricted sensor range, and narrow viewpoints, Co3SOP facilitates research into collaborative perception. The dataset provides dense and comprehensive occupancy annotations generated using a high-resolution semantic voxel sensor in the CARLA simulator, replaying existing collaborative perception scenarios.

Dataset Features

• High-Resolution Annotations: Provides a voxel-level representation of both geometric details and semantic categories.
• V2X Scenarios: Enables the exchange of information between multiple agents to enhance perception accuracy.
• Diverse Prediction Ranges: Establishes benchmarks with varying spatial extents (25.6m, 51.2m, and 76.8m) to assess the impact of range on collaborative prediction.

Citation

If you find this dataset or research useful, please consider citing:

@article{wu2025synthetic,
  title={A Synthetic Benchmark for Collaborative 3D Semantic Occupancy Prediction in V2X Autonomous Driving},
  author={Wu, Hanlin and Lin, Pengfei and Javanmardi, Ehsan and Bao, Naren and Qian, Bo and Si, Hao and Tsukada, Manabu},
  journal={arXiv preprint arXiv:2506.17004},
  year={2025}
}

Acknowledgements

This work builds upon several excellent open-source projects, including OpenCOOD, SurroundOcc, and LMSCNet.