DrivingStereo

This is a mirror of the original DrivingStereo download files, preserving the original ZIP archives and camera calibration text files without recompression. The original dataset was introduced at CVPR 2019 by Guorun Yang, Xiao Song, Chaoqing Huang, Zhidong Deng, Jianping Shi, and Bolei Zhou.

• Official website: https://drivingstereo-dataset.github.io/
• Paper: https://openaccess.thecvf.com/content_CVPR_2019/html/Yang_DrivingStereo_A_Large-Scale_Dataset_for_Stereo_Matching_in_Autonomous_Driving_CVPR_2019_paper.html
• Original license: https://github.com/drivingstereo-dataset/drivingstereo-dataset.github.io/blob/master/LICENSE

Contents

The inventory contains 300 source files totaling 202,346,258,264 bytes (202.346 GB). The early demonstration package is excluded; the released training, testing, weather, and calibration directories are mirrored in full.

Transfer status

Transfer in progress. As of 2026-10-17T03:37:54+00:00, 0/300 files (0.000/202.346 GB) are verified on the Hub. Status is refreshed after each batch. transfer_status.json contains the machine-readable status.

source_manifest.json records source Google Drive file IDs, URLs, original paths, expected byte sizes, mirror paths, and the approximately 50 GB batch plan. SHA256SUMS contains hashes for successfully uploaded files. Each ZIP is checked for CRC errors before upload; uploaded file sizes and content hashes are checked against Hub metadata before a local batch is removed.

Download

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="kronecker0122/DrivingStereo",
    repo_type="dataset",
    local_dir="DrivingStereo",
    allow_patterns=["train/left-image/*", "train/right-image/*", "train/disparity-map/*", "calibration/train/*"],
)

Download all files by omitting allow_patterns. ZIP archives must be extracted before using the original images and maps. Follow the official website for calibration details and the different disparity encodings at each resolution.

License and citation

The official website releases the dataset under the MIT license. The original copyright notice and license are included in LICENSE. Please cite the authors' publication when using the dataset:

@inproceedings{yang2019drivingstereo,
  title={DrivingStereo: A Large-Scale Dataset for Stereo Matching in Autonomous Driving Scenarios},
  author={Yang, Guorun and Song, Xiao and Huang, Chaoqin and Deng, Zhidong and Shi, Jianping and Zhou, Bolei},
  booktitle={IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2019}
}