Stripped Wire Dataset

This dataset belongs to the publication PB-IAD: Utilizing foundation models for semantic industrial anomaly detection in dynamic manufacturing environments. It contains images of individual electrical wires with four different diameters. The data is organized into separate subsets for training and testing to support anomaly detection and vision-language model (VLM) evaluation. Test set: Includes three labeled classes: Good – Wires without visible defects. Pulled strands – Wires with partially detached or displaced conductor strands. Cut strands – Wires with visibly damaged or severed strands. Training set: Divided into two subsets: PatchCore – Images representing the good class used for training a PatchCore-based anomaly detection model. VLM – A minimal set of three images, one representing each of the three classes (Good, Pulled strands, Cut strands), utilized for adapting or evaluating vision-language models. The dataset provides a controlled image collection focusing on wire-end conditions for experimental use in anomaly detection and visual classification tasks.