OpenStereo: A Comprehensive Benchmark for Stereo Matching

OpenStereo is a flexible and extensible project for stereo matching.

What's New

• [Sep 18, 2025]: Our paper makes public: StereoCarla: A High-Fidelity Driving Dataset for Generalizable Stereo.
• [June 12, 2025]: We have integrated the FoundationStereo model (training and inference). For details, please refer to prepare_foundationstereo.
• [Jan 28th, 2025]: The paper of LightStereo has been accepted by ICRA 2025.
• [June 26th, 2024]: TensorRT has been integrated, please see the Deployment documentation.
• [May 2024]: The 2.0 version of OpenStereo is available, featuring an optimized training and testing framework.
• [January 2024]: Our proposed StereoBase rank 1st on the KITTI15 leaderboard!!!
• [December 2023]: Our paper makes public: OpenStereo: A Comprehensive Benchmark for Stereo Matching and Strong Baseline.
• [March 2023]:OpenStereo is available!!!

Our Publications

• [Arxiv'25] StereoCarla: A High-Fidelity Driving Dataset for Generalizable Stereo, Paper, Data and ProjectPage.
• [ICRA25] LightStereo: Channel Boost Is All Your Need for Efficient 2D Cost Aggregation, Paper and Code.
• [Arxiv'24] Stereo Anything: Unifying Stereo Matching with Large-Scale Mixed Data, Paper, ProjectPage and Code.
• [Arxiv'23] OpenStereo: A Comprehensive Benchmark for Stereo Matching and Strong Baseline, Paper and Code.

Overall

Highlighted features

• Multiple Dataset supported: OpenStereo supports 17 popular stereo datasets: SceneFlow, KITTI12 & KITTI15, ETH3D, Middlebury, DrivingStereo, Sintel, FallingThings, InStereo2K,UnrealStereo4k, VirtualKitti2, CREStereo, Argoverse, Spring, TartanAir, FoundationStereo and StereoCarla.
• Multiple Models Support: We reproduced several SOTA methods and achieved the same or even better performance.
• DDP Support: The officially recommended Distributed Data Parallel (DDP) mode is used during both the training and testing phases.
• AMP Support: The Auto Mixed Precision (AMP) option is available.
• TensorRT Support: TensorRT has been integrated.
• Nice log: We use tensorboard and logging to log everything, which looks pretty.

Getting Started

Please see 0.get_started.md. We also provide the following tutorials for your reference:

• Prepare dataset
• Detailed configuration
• Customize model
• Advanced usages

Model Zoo

Results and models are available in the model zoo.

Acknowledgement

AANet   ACVNet   CascadeStereo   CFNet   COEX   DenseMatching   FADNet++   GwcNet   MSNet   PSMNet   RAFT   STTR   OpenGait   IGEV   NMRF   FoundationStereo   MonSter  

Citation

@article{OpenStereo,
        title={OpenStereo: A Comprehensive Benchmark for Stereo Matching and Strong Baseline},
        author={Guo, Xianda and Zhang, Chenming and Lu, Juntao  and Wang, Yiqi and Duan, Yiqun and Yang, Tian and Zhu, Zheng and Chen, Long},
        journal={arXiv preprint arXiv:2312.00343},
        year={2023}
}
@article{guo2024stereo,
        title={Stereo Anything: Unifying Zero-shot Stereo Matching with Large-Scale Mixed Data},
        author={Guo, Xianda and Zhang, Chenming and Zhang, Youmin and Wang, Ruilin and Nie, Dujun and Zheng, Wenzhao and Poggi, Matteo and Zhao, Hao and Ye, Mang and Zou, Qin and Chen, Long},
        journal={arXiv preprint arXiv:2411.14053},
        year={2024}
}
@inproceedings{guo2025lightstereo,
        title={Lightstereo: Channel boost is all you need for efficient 2d cost aggregation},
        author={Guo, Xianda and Zhang, Chenming and Zhang, Youmin and Zheng, Wenzhao and Nie, Dujun  and Chen, Long},
        booktitle={ICRA},
        year={2025}
}
@article{guo2025stereocarla,
      title={StereoCarla: A High-Fidelity Driving Dataset for Generalizable Stereo}, 
      author={Xianda Guo and Chenming Zhang and Ruilin Wang and Youmin Zhang and Wenzhao Zheng and Matteo Poggi and Hao Zhao and Qin Zou and Long Chen},
      year={2025},
      journal={arXiv preprint arXiv:2509.12683}
}

Note: This code is only used for academic purposes; people cannot use this code for anything that might be considered commercial use.