Concrete Crack Conglomerate Dataset

This dataset is the conglomeration of the cataloged crack datasets from the literature, making an extremely diverse crack dataset. There were over 10,995 images which have been merged from CFD (Shi, Cui, Qi, Meng, & Chen, 2016), Crack500 (Yang et al., 2020), CrackTree200 (Zou, Cao, Li, Mao, & Wang, 2012), DeepCrack (Y. Liu, Yao, Lu, Xie, & Li, 2019), Eugen Miller (Yang et al., 2020), GAPs (Eisenbach et al., 2017), Rissbilder (Yang et al., 2020), non-crack (Dorafshan, Thomas, & Maguire, 2018), and Volker (Yang et al., 2020). The majority of pixels were in the background class at 97.2%, and a little over 2.8% were in the crack class. This was expected, since cracks take up only thin lines of space on most images. The images were resized to 512x512 for training and testing the DeeplabV3+ model. The original and resized images are included. The data was split 90% training, and 10% testing by randomly sorting the data. After training with the DeeplabV3+ model (DOI: 10.7294/16628596), we were able to receive approximately 70% accuracy for detecting concrete cracks in the image. More details of the training, the results, the dataset, and the code may be referenced in the journal article. The GitHub repository information may be found in the journal article.

If you are using the dataset in your work, please include both the journal article and the dataset citation.