Crackseg9k Dataset Overview

Basic Information

• Dataset Name: crackseg9k
• Version: 4.0.0
• Homepage: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/EGIEBY
• License: Creative Commons CC0 1.0 Universal (CC0 1.0) Public Domain Dedication

Description

• This dataset is currently the largest, most diverse, and most consistent crack‑segmentation dataset.
• It includes 9 255 images combined from multiple smaller open‑source datasets.
• The dataset consists of 10 sub‑datasets: Crack500, Deepcrack, Sdnet, Cracktree, Gaps, Volker Rissbilder, Noncrack, Masonry, and Ceramic.
• All images have been pre‑processed and resized to 400 × 400 pixels.

Citation

@article{kulkarni2022crackseg9k, title={CrackSeg9k: A Collection and Benchmark for Crack Segmentation Datasets and Frameworks}, author={Kulkarni, Shreyas and Singh, Shreyas and Balakrishnan, Dhananjay and Sharma, Siddharth and Devunuri, Saipraneeth and Korlapati, Sai Chowdeswara Rao}, journal={arXiv preprint arXiv:2208.13054}, year={2022} }

Data Split

• Training set: data/train.parquet
• Test set: data/test.parquet

Features

• image: data type base64
• mask: data type base64
• head: data type base64