Abstract
VIADUCT data set provides 49 categories of industrial objects with corresponding 135 defect categories.
Technical Information
The data set consists exclusively of PNG files. Each category is divided into the sub-followers train, val, test and ground truth. Only defect-free images are in the train and val directory. In the test directory are the sub-folders for each defect category and the category "good" (defect-free). The annotations (also called ground truth or labels) for the images in the defect category directories are located in the ground truth directory mentioned above. The so-called anomaly maps are binary and mark with white on black pixels where the defect is in the image.