Procedural Dataset Generation for Zero-Shot Stereo Matching
Paper β’ 2504.16930 β’ Published
Error code: RowsPostProcessingError
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
WMGStereo is a procedural dataset generator specifically optimized for zero-shot stereo matching performance. This repository contains the WMGStereo-150k dataset, a large-scale synthetic training dataset featuring indoor, nature, and dense "flying" scenes.
You can download the dataset using the huggingface-cli:
pip install huggingface-cli
huggingface-cli download pvl-lab/WMGStereo --repo-type dataset
The dataset file structure is as follows:
.
βββ WMGStereo/
βββ indoor/
β βββ seed_num/
β βββ frames/
β βββ Image/
β β βββ camera_0
β β βββ camera_1
β βββ camview/
β β βββ camera_0
β β βββ camera_1
β βββ disparity/
β β βββ camera_0
β βββ occ_mask/
β β βββ camera_0
β βββ sky_mask/
β βββ camera_0
βββ flying/
β βββ ...
βββ nature/
βββ ...
.npz files that contain a dictionary with indices K, T, HW, corresponding to calibration, translation, and resolution matrices.If you find WMGStereo useful for your work, please consider citing the academic paper:
@misc{yan2025proceduraldatasetgenerationzeroshot,
title={What Makes Good Synthetic Training Data for Zero-Shot Stereo Matching?},
author={David Yan and Alexander Raistrick and Jia Deng},
year={2025},
eprint={2504.16930},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2504.16930},
}