Surface Quality Augmentation for Metalworking Industry with Pix2Pix

Image augmentation has become an important part of the data preprocessing pipeline, helping to acquire more samples by altering existing cutting, shifting, etc.. For some domains, augmenting images is not sufficient, due missing in domains (e.g., faulty work pieces or events that occur infrequently). In such a case, new must be generated, since with surface quality defects are often rare occurrence metalworking and amount even standard techniques does meet requirements train Convolutional Neural Network (CNN) for fault detection. This paper utilizes Pix2Pix image generate defects. The approach allows specifying kind defect, location, size transforms adding Furthermore, metrics evaluate augmented discussed recommendation best performing metric within domain given.