This dataset is similar to Common Objects inContext (COCO) and is generated for two major structural failure, cracks and spalling. A total of 2,229 images are prepared for training and validation. The images in this dataset are labeled by the tool referred as the COCO Annotator. Size of the trainingand labeling images varies from 147×288 to 4600×3070. Steel structures are not included in this dataset. But it inludes surface cracks on structural or nonstructural materials at various scales, and exposion of reinforcements in spalling case.
The training images and validation images needs to be merged into a train folder and a verification folder for the purpose of training and validation. And the labels for them are in train.json and validation.json, respectively.
The original images and JSON files used for training and validation are provided here, feel free to use them for your study, but please cite this as our contribution:
@Article{isprs-annals-V-2-2021-161-2021, AUTHOR = {Bai, Y. and Sezen, H. and Yilmaz, A.}, TITLE = {DETECTING CRACKS AND SPALLING AUTOMATICALLY IN EXTREME EVENTS BY END-TO-END DEEP LEARNING FRAMEWORKS}, JOURNAL = {ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences}, VOLUME = {V-2-2021}, YEAR = {2021}, PAGES = {161--168}, URL = {https://www.isprs-ann-photogramm-remote-sens-spatial-inf-sci.net/V-2-2021/161/2021/}, DOI = {10.5194/isprs-annals-V-2-2021-161-2021} }