Degree of delamination eval App

This dataset contains images of coated metal plates after a delamination process, where an extended delamination scratch is visible on the surface. For each sample, the dataset provides both the original image and a corresponding segmented mask. In the segmentation mask, the delaminated area is highlighted in white, while the rest of the image is represented in black. The dataset is organized into eight main folders, each containing two subfolders named inputs and targets. The inputs subfolder contains the original or augmented images of the coated metal plates, and the targets subfolder contains the corresponding binary segmentation masks. Folder 1 contains only the original dataset, consisting of 456 images and their corresponding masks. The remaining seven folders contain progressively expanded versions of the original dataset, where artificial images have been added to the original set. In these folders, the number of artificially generated images corresponds to multiples of the original dataset size, specifically 0.2×, 0.5×, 1×, 2×, 3×, 5×, and 10× the number of original images. This structure enables direct comparison between the baseline dataset and multiple levels of dataset augmentation, making it suitable for training and evaluating image segmentation models focused on detecting delamination areas on coated metal surfaces.