"A Real World Textile AI Dataset for Textile Quality Inspection"

"The resulting dataset is comprised of digital images of fabrics. Images are acquired using industrial camera. Specification of camera is given as Machine Vision Camera Model No - VCXG.2-57C and Machine Vision Lens Size - 6mm M118FM16. It also has MACHINE VISION RING LIGHT Model No - CMVRL1366925240210 and Machine Vision Camera Enclosure Material and camera enclosure mounting bracket. The device is mounted on Setup furniture including mountable table and lights.  For model training, images from these directories were loaded and processed using ImageDataGenerator with a target size of 512x512 pixels and in batches of 32. The class_mode for data loading was set to 'categorical', and the class labels used for classification were derived directly from the directory structure of the training data. The dataset was designed to encompass various fabric faults such as broken pick, hole, missing pick, thick weft and perfect deformations. This project involved the systematic collection and annotation of fabric images to create a comprehensive dataset for fabric fault detection. The data collection process commenced with gathering a diverse dataset of fabric images, which included both defect-free samples and defective samples of various types and severities. Following collection, these images were meticulously annotated to clearly indicate defect locations and their specific types, which served as the fundamental input for subsequent supervised learning tasks.The reuse potential of this dataset and the system developed upon it is significant. The system is engineered to be scalable and adaptable to diverse fabric types, suggesting its applicability across different textile products and manufacturing lines. Its design facilitates easy integration into existing textile manufacturing processes, thereby improving quality control and mitigating human errors commonly associated with manual inspection methods. Furthermore, the dataset provides a robust foundation for future extensions of the system, such as incorporating the ability to detect other fabric faults. The project also envisions supporting real-time detection and classification of faults by integrating the system with cameras and sensors on textile production lines. This potential for broader application and continued development underscores the dataset's utility in advancing automated textile inspection.[1] Do not use words such as \u2018study\u2019, \u2018results\u2019, or \u2018conclusions\u2019 \u2013 a data article should only describe your data. Also, please ensure that the abstract differs from the description provided in the data repository."