CHCrack5K Dataset

A Comprehensive Dataset for Pixel-level Crack Detection

Overview

CHCrack5K is a robust dataset designed for advanced crack detection research. It combines 11 publicly available crack datasets into a unified set, resulting in 5,014 labeled image samples. Each dataset undergoes specific preprocessing to standardize all samples to a resolution of 480×480 pixels. This dataset offers a diverse range of crack structures, providing a more challenging and realistic benchmark for testing robust crack detection algorithms.

Dataset Details and Processing

Each of the 11 contributing datasets has been processed to ensure consistent dimensions and quality:

CFD, MCD-Crack, and Road420: Padding applied to retain spatial structure.
Crack500 and SUT-Crack: High-resolution images resized and cropped.
DeepCrack537: Padding and resizing applied for uniform dimensions.
CrackTree260 and other datasets: Resized with necessary adjustments to standardize dimensions.

Dataset Composition

The dataset is divided as follows:

Training set: 3,510 samples
Validation set: 501 samples
Testing set: 1,003 samples

Table I: Overview of the CHCrack5K Dataset

Dataset	Original Resolution	Post-Processing	Samples
CFD	480×320	Padding	117
Crack500	2560×1440/...	Cut + Resize	1,000
Road420	448×480	Padding	420
MCD-Crack	480×320	Padding	235
SUT-Crack	3024×4032	Cut + Resize	260
CamCrack789	640×640	Resize	789
DeepCrack537	544×384	Padding + Resize	537
CrackTree260	800×600	Padding + Resize	260
CrackLS315	512×512	Resize	315
CRKWH100	512×512	Resize	100
TunnelCrack	512×512	Resize	981

Access the Dataset

The dataset is hosted on GitHub and can be accessed via the following link:

Download CHCrack5K Dataset

Citation

If you use CHCrack5K in your research, please cite this dataset as follows: @dataset{CHCrack5K, author = {Your Name}, title = {CHCrack5K: A Comprehensive Crack Detection Dataset}, year = {2024}, url = {https://github.com/hanshenchen/CHCrack5K} }