Shearographic Anomaly Detection Dataset (SADD)

This dataset contains shearographic measurement data of simulated kissing bonds in a polymer–rubber adhesive joint. The defects were created by inserting defined discontinuities within the adhesive layer, producing fully detached bonding regions. In addition, the dataset includes reference measurements without defects, which contain only global deformation used to generate the phase jumps. The dataset consists of three classes of samples: Defective samples with simulated kissing-bond anomalies (Faulty) Non-defective samples without any adhesive discontinuities and without global Deformation (Good_clean) Non-defective samples without any adhesive discontinuities and with global deformation (Good_stripes) Ground-truth annotations (Bounding Box, CSV and YOLO Format) are provided only for the defective regions in the defective samples (Faulty). The background and all non-defective areas remain unlabeled. The labels were manually created by a domain expert. This results in partially labeled data that may be used in supervised, weakly supervised, or unsupervised analysis methods.