Key points are not available for this paper at this time.
The dataset includes 458 hi-res images together with their alpha maps (BW) indicating the crack presence. The ground truth for semantic segmentation has two classes to conduct binary pixelwise classification. The photos are captured in various buildings located in Middle East Technical University. You can access a larger dataset containing images with 227x227 px dimensions for classification which are produced from this dataset from http://dx.doi.org/10.17632/5y9wdsg2zt.1 .
Çağlar Fırat Özgenel (Wed,) studied this question.