CoST: Efficient Collaborative Perception From Unified Spatiotemporal Perspective

This is the official implementation of the ICCV 2025 Highlight paper
"CoST: Efficient Collaborative Perception From Unified Spatiotemporal Perspective"
by Zongheng Tang, Yi Liu, Yifan Sun, Yulu Gao, Jinyu Chen, Runsheng Xu, and Si Liu.

📄 Paper: arXiv:2508.00359
🎥 Introduction Video: YouTube

---

✨ Highlights

• Unified Spatiotemporal Perspective:
  CoST introduces an efficient framework that jointly models spatial and temporal collaboration among connected agents.
• Efficiency-Oriented Design:
  Achieves high perception accuracy with reduced communication cost and computation overhead.
• Multi-dataset Validation:
  Verified on V2V4Real, DAIR-V2X, and V2XSet datasets, demonstrating strong generalization ability across scenarios.

---

📦 Supported Datasets

• V2V4Real
• DAIR-V2X
• V2XSet

📁 Data Download and Structure

Please check our website to download the data (OPV2V format).
After downloading, organize the data as follows:

├── v2v4real
│   ├── train
│   │   ├── testoutput_CAV_data_2022-03-15-09-54-40_1
│   ├── validate
│   ├── test

---

🧰 Devkit Setup

CoST’s codebase is built upon V2V4Real.

Follow these steps to set up the environment:

1. Create Conda Environment (Python ≥ 3.7)

conda create -n v2v4real python=3.7
conda activate v2v4real

2. PyTorch Installation (≥ 1.12.0 Required)

Example for CUDA 11.3:

conda install pytorch==1.12.0 torchvision==0.13.0 cudatoolkit=11.3 -c pytorch -c conda-forge

3. spconv 2.x Installation

pip install spconv-cu113

4. Install Other Dependencies

pip install -r requirements.txt
python setup.py develop

5. Build CUDA Extension for Bounding Box NMS

python opencood/utils/setup.py build_ext --inplace

6. Install Deformable Convolution

cd opencood/models/sub_modules/ops
sh make.sh

---

🚀 Quick Start

• Training:
  Run the provided training script

  bash train.sh

• Testing:
  Run the provided testing script

  bash test.sh

---

📜 License

This project is released under the Apache 2.0 License.
See the LICENSE file for details.

---

📚 Citation

If you find CoST useful in your research, please cite:

@article{tang2025cost,
  title   = {CoST: Efficient Collaborative Perception From Unified Spatiotemporal Perspective},
  author  = {Tang, Zongheng and Liu, Yi and Sun, Yifan and Gao, Yulu and Chen, Jinyu and Xu, Runsheng and Liu, Si},
  journal = {arXiv preprint arXiv:2508.00359},
  year    = {2025}
}