Datasets:
license: mit
pretty_name: DrivingStereo
task_categories:
- depth-estimation
tags:
- stereo-matching
- autonomous-driving
- computer-vision
size_categories:
- 100K<n<1M
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
| Directory | Contents |
|---|---|
train/ |
All 38 training sequences: half-resolution left/right images, disparity maps, and depth maps |
test/half-size/ |
All four testing sequences at half resolution, with all four modalities |
test/full-size/ |
All four testing sequences at full resolution, with all four modalities |
weather/ |
Sunny, cloudy, foggy, and rainy subsets, with both released resolutions and all four modalities |
calibration/ |
Training and testing camera calibration, for both resolutions |
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 2026-10-24T01:50:07+00:00, 213/300 files
(85.182/202.346 GB) are verified on the Hub.
Status is refreshed after each batch. transfer_status.json contains the machine-readable status.
source_manifest.json records original Google Drive file IDs and URLs, official
Baidu share links, original paths, expected byte sizes, mirror paths, and the
approximately 50 GB batch plan. Files may be retrieved from either official
mirror; a Baidu file must match the original filename and exact byte size.
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}
}