This dataset reports the results of applying a new method that combines the block based and pixel based approaches for crack detection as described in 

Mohamed Abdellatif, Harriet Peel, Anthony G. Cohn and Raul Fuentes, Combining Block-based and Pixel-based
Approaches to Improve Crack Detection and Localisation, Automation in Construction, 103492, 2020.

The dataset reports the method performance ( called Combined or Comb) running on three standard crack datasets ( Crack Forest (CF), CRACK500 and CrackIT ) and compares the performance with respect to three deep learning methods namely, HED, RCF and FPHB.

The Folders are arranged per dataset name.
-CF
-CRACK500
-CrackIT

Each folder contains a set of 9 folders as follows:
-Original images
-Ground truth images
-Results of Combined method
-Results of HED method
-Results of RCF method
-Results of FPHB method
-Results of HED method compared with ground truth
-Results of RCF method compared with ground truth
-Results of FPHB method compared with ground truth

The colour code for comparison with ground truth is as follows:
White: True  Positive 
Red  : False Positive
Cyan : False Negative