Structural Defects Dataset (S2DS)

The S2DS dataset (<u>S</u>tructural <u>D</u>efects <u>D</u>ata<u>s</u>et) features 743 images of mostly concrete surfaces representing typical structural defects – such as crack or spalling – or control points. Each image is a 1024×1024 patch from a larger image. The classes covered are background, crack, spalling, corrosion, efflorescence, vegetation, control point.

Terms and Conditions of Use

The S2DS dataset is publicly available for academic use only. While every effort has been made to ensure the accuracy of this database, we cannot accept responsibility for errors or omissions. The academic use of this database is free of charge. Any commercial distribution or act related to the commercial usage of this database is strictly prohibited. The distribution of this database to any parties that have not read and agreed to the terms and conditions of usage is strictly prohibited. Neither Bauhaus-Universität Weimar, nor any third parties who may provide information to us for the dissemination purpose, shall have any responsibility for or be liable in respect of the content or the accuracy of the provided information, or for any errors or omissions therein. The Bauhaus-Universität Weimar reserves the right to revise, amend, alter or delete the information provided herein at any time, but shall not be responsible for or liable in respect of any such revisions, amendments, alterations or deletions.<br> _{(Adapted from GAPs request form, https://www.tu-ilmenau.de/fileadmin/Bereiche/IA/neurob/Datasets/Request-Gaps3.pdf.)}

Citation

The dataset must not be shared or distributed.

Always link and cite the work properly:

@inproceedings{benz2022image,
  title={Image-Based Detection of Structural Defects Using Hierarchical Multi-scale Attention},
  author={Benz, Christian and Rodehorst, Volker},
  booktitle={DAGM German Conference on Pattern Recognition},
  pages={337--353},
  year={2022},
  organization={Springer}
}

Download

... from Google Drive

The S2DS dataset can be downloaded from Google drive:

https://drive.google.com/file/d/1PQ50QKfy2vnDOHSmw5bpBFi33hZsSXuM/view?usp=sharing

<!--- ### ... using Git *(only test set)* The S2DS **test set** was pushed to github using Git LFS (*large file storage*, https://git-lfs.github.com/). It can be obtained running the usual: ``` git clone git@github.com:uncle-ben-z/s2ds.git ``` -->

Classes

| Class | Description | Color | |---------------|----------------------------------------------|--------------------------------------------------| | Crack | Linear fractures in material | | | Spalling | Material detachment from surface | | | Corrosion | Rust formation by oxidizing metal parts | | | Efflorescence | Depositions of dissolved chemicals on surface | | | Vegetation | Surficial plant growth | | | Control Point | Geodetic fiducial markers for georeferencing | | | Background | All other classes | |

Split

The data were manually split into training, validation, and test set. Images of distinctly poor quality or almost invisible defects were incorporated into the training (and validation) set. The test contains only images with a fair chance of recognition.

| | Training | | Validation | | Test | | |---------------|----------|---|------------|---|--------|--------| | Class | Images | Pixels | Images | Pixels | Images | Pixels | | Crack | 180 | 0.6 M | 25 | 0.1 M | 27 | 0.1 M | | Spalling | 151 | 39.2 M | 23 | 3.3 M | 20 | 4.2 M | | Corrosion | 209 | 8.8 M | 36 | 0.5 M | 38 | 0.9 M | | Efflorescence | 96 | 4.6 M | 13 | 0.6 M | 17 | 1.5 M | | Vegetation | 97 | 15.7 M | 16 | 2.7 M | 18 | 2.9 M | | Control Point | 70 | 1.5 M | 9 | 0.2 M | 10 | 0.2 M | | Background | 556 | 519.9 M | 87 | 83.8 M | 93 | 87.7 M | | Total | 563 | 590.3 M | 87 | 91.23 M | 93 | 97.52 M|

Examples

Downscales images from the test set.

| Image | Label / Annotation | | Image | Label / Annotation | |--------------------------------------|---------------------------------|----|--------------------------------------|----------------------------------------| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | ![Image 1613](thumbnails/1613_lab